Moving object tracking method and system thereof

Technical Field

The present invention relates to video monitoring technology, in particular to a kind of intelligent video monitor system target tracking movement of the method and its system.

Background Art

With the increase in criminal horizontal and threats, security has become an issue of concern throughout the world. Video monitoring is to solve the problem of the method. In addition to public safety, video monitoring can also be effective in resolving some other problems, such as a crowded city traffic, the regulation of the flow of people. For many years such as large-sized monitor system in the airport, bank, highway or urban centres the site of such as has been widely applied.

The traditional video monitoring technology is generally manual monitoring, the presence of fatigue, easy to negligence, slow reaction speed, deficiencies of manual cost is high. Therefore, in recent years there has been a gradual research a digitized, standardized, intelligent and IP network intelligent video monitoring technology.

Conventional intelligent video monitoring technology includes a moving target tracking technology. The purpose of tracking a moving target in the correct is detected on the basis of the moving target, with the goal of determining the position of the different process in the Image scene.

In order to realize the tracking, can be used for analysis of motion-based, such as inter-frame difference method and flow dividing method. Adjacent frame interframe difference method is the Image after the subtraction operation, the threshold value and dividing the result Image, extracting a moving object. The disadvantages of this method is it can only be according to the interframe pixel intensity changes to detect whether the object in the scene motion, movement of the target signal and the noise of the interframe correlation of interframe correlation are very weak, it is difficult to distinguish. Flow dividing method is through the target and the background to detect different speed between the moving target. The disadvantages of this method is not able to effectively distinguish from the background of the target caused by a movement of, the aperture of the display and, large amount of calculation, require special hardware support.

In order to realize the tracking, Image matching method can be used, such as the area matching, model matching. The region matching of a block of the reference Image with a real-time Image of all possible position of the superimposed on, then calculates a corresponding Image similarity measure, its maximum similarity is a position corresponding to a position of the object. The disadvantages of this method is the large amount of calculation, it is difficult to achieve real-time requirements. Model matching is based on the template to be matched in the target Image scene. The disadvantages of this method is to calculate and analyse the complex, the operation speed is slow, updating of the model is more complex, real-time performance is poor.

To sum up, there is an urgent need for more simple, effective, real-time moving target tracking scheme.

Content of the invention

In view of this, the main purpose of this invention is to provide a moving target tracking method and its system, which can obtain the correct foreground Image, target detection errors is reduced; further, can be carried out based on the detection results of the prediction, matching, update operation, in order to filter out false moving target, to achieve accurate tracking of a moving target.

In order to achieve the above-mentioned purpose, the technical scheme of the present invention is realized as follows:

The present invention provides a method for tracking a moving target, the moving target tracking method comprises:

Detecting target, the moving target in the video scene from the background region;

Predict a target, estimating the target of movement for the next frame;

Matching target, tracking matching the stable target, and filtering the false target;

Updated target, updating the template stability objectives the current frame.

According to the present invention, the detection target comprises the following steps:

Obtaining video, acquire video content in order to get the scene Image, and the establishment of the background model;

Pre-processing the Image, the elimination of scene Image the impact of the background model;

Mark region, according to the background model for foreground segmentation of the scene Image, and mark the connected region;

Maintaining a state, determine a detection target module the current state, to make the corresponding processing, and the abnormality detection is necessary;

Enhancing regional, the use of shadow detection, high brightness detection and tree filter, eliminating shadow, high brightness and false region of the leaves;

Separatism and merged regions, using background model provides the restraint of the a priori knowledge of the people and vehicle model and the regional are combined and split and process, in order to solve the objectives and excessive segmentation on each other.

Wherein, the pre-processing Image includes: filtering processing and the overall motion compensation; wherein,

The filter processing includes: a noise filter to the Image processing, Image smoothing processing;

The global motion compensation, because the camera is to compensate the Image caused by the slight swinging movement of the whole, in the global motion compensation, the motion model comprises a translation, rotation, zoom.

Prospects for by the following conventional formula of the rectangular area where positive and negative around 5 luminance difference between the area of a picture IDS, to obtain global motion compensation Δx translation distance of the Image in the, Δy, formula is as follows:

IDS = &# X003A3; x = s x m &# X003A3; y = s y n ( I ( x , y ) ( t ) - I ( x , y ) ( t - 1 ) ) s x s y

wherein, s_x representing region starting point coordinate x, s_y representing region starting point coordinate y, I_(x,y) (t) said current frame Image gradation, I_(x,y) (t-1) of Image gray scale expressed previous frame; calculating the position of the four other area variation, the average of the last determined Δx, Δy; Δx the Image in accordance with the, after compensation to translate Δy obtained Image.

Wherein, said mark region includes the following steps:

Foreground segmentation, the background model-based Image segmentation of the scene, in order to get the prospect of the binary Image;

Morphological processing, using the mathematical morphology of the binary method for the treatment, in order to remove false region of small area, and the area of the filling area is relatively large; and

Connected region mark, the connection area for the same method of the different region of the scene, in order to distinguish different target area.

Wherein, the maintenance state includes a state decision and abnormality detection.

The state determination, detecting target module is to judge the state of the current, to make the corresponding processing; stable time the scene on the spot exceeds the threshold 1, system to enter into the working state by initializing a state; changing time the scene on the spot exceeds the threshold value 2, the system is composed of a working state of the initialization state. Wherein, the threshold value 1 is preferably 0.5-2 seconds, the threshold 2 preferably 5-20 second.

The abnormal detection, is serious disturbance of the video signal, and of some people are carried out when the condition of the camera; the edge of the two time according to the matching value of the background and the background initialization successful the shortest time judging, if the background of the current frame the edge of the background model is less than a threshold value matching 3 or background initialization successful exceeds a threshold value of 4 the most short period of time, are considered to be abnormal phenomenon. Wherein, the threshold value 3 is preferably 30-50 between, the threshold value 4 is preferably 6-20 second.

Wherein, the stiffening region includes: shadow detection, high brightness detection, tree filtering.

Shadow detection, is directed to each of the connected region, separately calculating the mean value of the pixel values in the region, and the average value as a threshold value, it is determined that the area of shadow region, then filtering the shadow region, if the pixel value is less than the threshold value, it is judged that the shadow.

High brightness detection, which is used for detecting whether the picture is in a high brightness state, if so, luminance compensation is performed, the brightness compensation so that the Image pixel values of the mean value of 128.

Tree detecting, in the Image is used for detecting the swinging leaves and swinging leaves shadow, and will be filtered from the foreground Image.

Swing leaf detection is based on the following two features of determining one of : (1) movement trajectory tracking, when the movement of the goal in path spot area of a region corresponding to the region of movement of the movement portion is less than the threshold value of the area of a region 5 time, the goal is that the swinging leaves ; (2) center-of-mass motion amplitude, when the adjacent path spot displacement of the mass center of the target in a target variation exceeds the threshold value of the width of the 6 time, the goal is that the swinging of the leaves. Wherein, the threshold value 5 is preferably 5%-15% between ; the threshold 6 is preferably 1.5-2.5 between.

Swing leaf shadow detection method is: statistical expansion operation respectively before and after the expansion operation in the region before and after the value of the pixels " 1 the number of the points of [...] , and calculating a ratio of them, if the ratio is less than the threshold value 7, that the shadow of this area is the area of the swinging leaves. Wherein, the said threshold 7 is preferably 40%-60% between.

Wherein, said split and merged regions the enhanced region is based on the processing procedure, it is judged whether two adjacent regions of the same target area; if it belongs to the same target area, then the two of the region merging; otherwise, the split thereof; wherein, two adjacent region is less than the threshold distance to the edge region 8 of the regional. Wherein, the said threshold 8 preferably 3-7 between pixels.

According to the present invention, the predicted target is based on the cumulative displacement of target movement and its corresponding accumulation time, calculating the average of the target moving speed, and on the basis of the speed prediction goal of the next displacement; wherein,

The accumulated displacement, and accumulating time in relationship to the average moving speed:

V=s/t

Wherein, the target s movement of the mass center of the stable displacement of the multiframe, the goal of the multiframe t the time required for the movement, for the v the average velocity of the stable target movement;

Predicting v according to the average speed of the next displacement is:

S ' =v·Δt

Wherein, the target time to predict Δt, s ' Δ t the goal of stable movement of the mass center of the displacement of the time.

According to the present invention, the matching target including: tracking matching the stable target and filtering the false target; wherein,

The stability of the tracking matching objective is to determine a detection region is matched with the target tracking, the matching according to the following formula with the target detection area in the matching coefficient to judge D:

D=Da*A_Da +Db*A_Db +Dc*A_Dc

Wherein, an area matching coefficient Da, histogram matching coefficient for Db, Dc a distance matching coefficient. When the detecting region D the target matching coefficient is larger than the threshold 9, the, it is judged that the detection region is matched with the target. A_Da, A_Db, A_Dc are Da, Db, the weight coefficient corresponding to Dc. Wherein, the said threshold 9 is preferably 0.7-0.8 between.

Area matched coefficient Da, when the detection area is the target area that intersects the target area is greater than the threshold value of the area of the 10 time, the detection region that satisfies area matching, the Da 1 ; otherwise, the Da 0. Wherein, the threshold 10 is preferably 40%-60% between.

The histogram matching coefficient Db, the goal is that when the detecting region of intersection is greater than a target histogram of area 11 of the threshold value of the histogram at the time, to meet the detection region that the matching of the histogram, taking Db 1 ; otherwise Db heating 0. Wherein, the threshold value 11 is preferably 40%-60% between.

Distance match coefficient Dc, according to the detection region is stationary is also of the movement of the two situations to consider distance match coefficient Dc; if the current frame in the previous frame of Image sensing area in the difference Image, the number of the former spots background spot 12 is larger than the threshold value of the number is, the movement of the detection area is that, otherwise that the detection area is stationary.

When the detection area is the movement, calculate the current frame of Image detecting the center of the area to be detected with the current frame the distance of the center of the sensing area, if the distance is less than the target where the length of the diagonal line of the rectangular frame 13 of the threshold value, the distance that the matching of the meeting, taking Dc 1 ; otherwise Dc heating 0.

When the detection area is stationary, the front one frame of the Image calculated detecting the center of the area to be detected with the current frame the distance of the center of the sensing area, if the distance is less than the threshold value 14, the distance that the matching of the meeting, taking Dc 1 ; otherwise Dc heating 0.

Wherein, the said threshold 12 is preferably 65%-75% between. The threshold 13 is preferably 1.5-2 between. The threshold value 14 is preferably 8-12 between pixels.

Filtering the false target is the track of the target motion analysis, in order to filter out false target area; wherein, the track analysis is to utilize the target trajectory information, statistical area change the smoothness and center-of-mass point of the stability of the change.

The statistics refer to the area change of the smoothness of the statistical target trajectory of the area set point {area₁, area₂, ... , area_n}, n that the number of the track points, mean value of statistical area:

Area [! OverBar! ] = 1 n &# X003A3; i = 1 n Area i

Statistical area variance: Area Sd = 1 n &# X003A3; i = 1 n ( Area i - Area [! OverBar! ] ) 2

When the area_sd/area> 0.5 time, that area does not change smoothly, the filtering the target area;

The statistical center-of-mass point changes in accordance with the normal objective of the stability of the movement is in the direction of will not produce recurrent mutation, in statistical adjacent path spot the ratio of the change of direction, if the ratio exceeds the threshold 15, the center-of-mass point is changed is not stable, the filtering the target area. Wherein, the threshold value 15 is preferably 40%-60% between.

According to another aspect of the invention, the invention also provides a moving target tracking system, the moving target tracking system comprises:

Detecting target module, is used for the movement of the video scene in the Image of the target area from the background;

Predict a target module, the moving target is used for estimating the next frame the position of the scene Image;

Matching the target module, the stability of the matching used for tracking the target, and filtering the false target; and

Updating the target module, for updating the template stability objectives the current frame.

Wherein, the detecting object module includes:

Obtaining the video module, is used for obtaining the video content in order to get the scene Image, and the establishment of the background model;

Pre-processing the Image module, the Image of the scene is used for eliminating the impact of the background model;

Mark zone module, according to the background model used for foreground segmentation of the scene Image, and mark the connected region;

Maintaining a state module, detecting the target module for determining the state of the current, to make the corresponding processing, and the abnormality detection is necessary;

Enhancing regional module, used for using shadow detection, high brightness detection and tree filter, eliminating shadow, high brightness and leaves swing of false area; and

Splittist and integration of regional module, used for using the background model provides the restraint of the a priori knowledge of the people and vehicle model and the regional are combined and split and process, in order to solve the objectives and excessive segmentation on each other.

The matching goal module comprises: tracking matching the stable target module, is used for judging the detection region is matched with the target tracking; and filtering the false target module, used for filtering the false area.

The biggest advantage of the present invention under complex background lies in realizing the accurate tracking of a plurality of objects, solves the problem that the weatherstripping, leaves the question swinging, and operation is simple, it has very strong practicability

The advantages of the invention also lies in the accurate detection of the moving object in the Image scene, include a person, vehicle, picture jitter can be neglected at the same time, the swing of the tree, the brightness variations, shadow, rain, snow, the impact of the interference factors.

The invention can also be used for intelligent video monitoring system, in order to achieve the target classification identification, alert moving target, moving target tracking, the tracking PTZ, automatic close-up shooting, target behavior detection, flow detection, congestion detection, carry-over detecting, is robbed the thing detection, smoke detection and flame detection functions.

Description of drawings

Figure 1 is a schematic structure of the target tracking method;

Figure 2 the method for tracking the moving target of this invention in the flow schematic diagram of the detection target;

Figure 3 of this invention method for tracking a moving target in the flow schematic diagram of the marker region;

Figure 4 is the moving target tracking method of the present invention in flow schematic diagram of the matching target;

Figure 5 is a schematic diagram of the structure of the target tracking system;

Figure 6 the moving target tracking system of the present invention in detecting target module structure schematic;

Figure 7 the movement of the target tracking system of the present invention in the structure of the matching the target module schematic diagram.

Mode of execution

Figure 1 of the present invention shown in the schematic structure of the target tracking method, as shown in Figure 1, moving target tracking method comprises:

Detecting target 10, the movement of the video scene in the target area from the background;

Predict a target 20, the estimated movement of the goal of a frame;

Matching target 30, the stability of the tracking matching target, and filtering the false target;

Updated target 40, updating the template stability objectives the current frame.

1st step first detecting target, the movement of the in the video scene to the target area from the background. Figure 2 to the invention in schematic the framework of the detection target, as shown in Figure 2. Detecting target 10 comprises a schematic diagram of the framework of: obtaining the video 11: acquire video content in order to get the scene Image, and the establishment of the background model; pre-processes the Image 12: elimination of scene Image the impact of the background model; mark region 13: according to the background model for foreground segmentation of the scene Image, and mark the connected region; maintaining a state 14: determine a detection target module the current state, to make the corresponding processing, and the abnormality detection is necessary; reinforced area 15, the use of shadow detection, high brightness detection and tree filter, eliminating shadow, high brightness and leaves false area of the; division and merged regions 16, the restraint of the use of background model provides a priori knowledge of a man and vehicle model and the region of a joint and splittist processing, in order to solve the objectives and excessive segmentation on each other.

First acquiring video 11 is through the content of the video capture device, the video capture device can be a visible spectrum, near-infrared or infrared video camera. The near infrared and infrared camera allows the light of the light without additional application. The building the background model initially in 1st frame scene Image as a background model, in the maintenance state after 14 to update in.

Then pre-processing Image 12 comprises a filtering processing and the global motion compensation. The filter processing means the Image as noise filtering, smoothing and other conventional processing, in order to remove the noise point in the Image. Filter treatment can be realized by the following literature, such as :" Image denoising mixed filter method. China's Image graphics Journal, 2005, 10 (3) "," improvement of the self-adaptive center weighted mean value filtering algorithm. The Tsinghua University Journal (natural scientific version), 1999, 39 (9) ".

Global motion compensation is to compensate the slight swing of the camera Image caused by movement of the whole. In the global motion compensation, the motion model are basically reflect various movement of the camera, comprises a translation, rotation, zoom, and the like. Method of global motion compensation is: region-based motion compensation block matching, draw in an Image four regional block, the length and width of the regional block 32-64 between pixels, require area covering relatively fixed background, such as a building, or stationary background.

The conventional global motion compensation method is as follows: assume that prospects for the size of the rectangular area where m×n, around the area calculated positive and negative 5 luminance difference between the area of a picture IDS, formula is as follows:

IDS = &# X003A3; x = s x m &# X003A3; y = s y n ( I ( x , y ) ( t ) - I ( x , y ) ( t - 1 ) ) s x s y

wherein, s_x representing region starting point coordinate x, s_y representing region starting point coordinate y, I_(x,y) (t) said current frame Image gradation, I_(x,y) (t-1) of Image gray scale expressed previous frame.

In this way the minimum luminance difference can be obtained corresponding to the position of the region, the position variable quantity of calculating this area Δx, Δy. Calculating the position of the four other area variation, the average of the last determined Δx, Δy. In accordance with the picture Δx, after compensation to translate Δy obtained Image.

Figure 3 the marker region in the present invention 13 flow schematic, as shown in Figure 3. Regional mark 13 specific flow of the procedure is as follows: the prospect segmentation 131, morphology processing 132, connected region mark 133.

The prospect segmentation 131 refers to the background model-based Image segmentation of the scene, in order to get the prospect of the binary. In particular, the scene Image and the background model correspond to the pixel values of the subtraction, if the result is greater than the set threshold value, it is recorded as "1 the front to the tourist attractions [...] ; if less than the threshold value, denoted as the" 0 the background spot[...] to the, the prospects for the binary Image.

Morphology processing 132 refers to the use of mathematical morphology of the binary method for the treatment, that is, through first expansion after corrosion, processing of the binary Image, in order to remove false region of small area, and the area of the filling area is relatively large. Wherein, corrosion parameter selected is 3 × 3 templates, expansion parameter selected is 3 × 3 templates.

Connected region mark 133 of the unit is usually means the same method of the different region of the scene, in order to distinguish different target area. Regional marking method can be communicated through the four or eight valuly not valuly not method for method to realize. Balian/four valuly not method of marking is of the: first of all, the morphology processing 132 line-by-line scanning operation of the captured Image, to find a 1st point of the marker region, mark the point; point for the inspection of the NCCs is a balian/four and marked meet the requirements of the connectivity, the mark is not yet point, at the same time, the additional marking point recorded as "region growing" seed point. In the labeling process in the follow-up of, continuously from the recording seed point in the array of a seed, the operation of the above-mentioned purposes, the oil is circulated in this way, until the recording of seed points of the array is empty, the end of a connected region mark. Then mark the area of mark a, until the morphology processing 132 of all of the captured Image is communicated with the mark region.

The area of the mark 13 in, with a single single area one-to-one corresponding to the target is not. Because the shielding situation, a regional comprising a plurality of individual drainability; because the background is similar to that for, a target may be excessive split into a plurality of areas; the impact of the illumination, may be included in area of shadow and highlighted area; the movement of the some of the non-interest, such as tree leaves swing and water wave poppling, and the like, will also produce false foreground region. These problems are inherent in the background model method, need to be addressed in the subsequent step.

Figure 2 in maintaining a state 14 including: state determination and abnormality detection.

State determination means determine a detection target module the current state, and make the corresponding processing. Detecting target module determines the state of the current mainly through the scene stabilization time, to judge the time of scene changes. Stable time the scene on the spot exceeds the threshold 1, system to enter into the working state by initializing a state; changing time the scene on the spot exceeds the threshold value 2, the system is composed of a working state of the initialization state.

The threshold value 1 is preferably 0.5-2 seconds. The threshold 2 preferably 5-20 second.

When in the working state, to continue to carry out the next operation, the background model is not changed. When the is in the initialization state, the re-establishment of the background model, and the abnormality detection is to be made when necessary. During the re-establishment of the background model, can be carried out through the inter-frame difference method for the detecting area. Interframe difference Image is taken through two of the absolute value of the subtraction.

Abnormality detection, including the necessary video signal interference is serious, a camera, the implementation of the situation. According to two times the edge of the background and the background initialization successful matching value of judging the shortest period of time. If the current frame the edge of the background and the background model is less than the threshold value the value of the match 3 or background initialization successful exceeds a threshold value of 4 the most short period of time, are considered to be abnormal phenomenon.

The threshold value 3 is preferably 30-50 between. The threshold value 4 is preferably 6-20 second.

Figure 2 the regional enhanced 15, is to use shadow detection, high brightness detection and tree filter, eliminating shadow, high brightness and leaves swing of false area; includes: shadow detection, high brightness detection, tree filtering.

Prospects for detecting the shadow detection of shadow region in the Image, including human, the shadow of the vehicle, and the filtering of shadow region is detected. The shadow detection is directed to each of the connected region, separately calculating the mean value of the pixel values in the region, and the average value as a threshold value, it is determined that the area of shadow region, then the shadow region filtering. Shadow decision rules are as follows: if the pixel value is less than the threshold value, it is judged that the shadow.

High brightness detection for detecting whether the picture is in a high brightness state (high brightness state means in the Image generally too high pixel value), if brightness compensation is carried out. Brightness compensation is realized by luminance balance, so that the Image pixel values of the mean value of 128.

Tree filtering is used for detecting the swinging of the leaves in the Image and its shadow, and will be filtered from the foreground Image.

Detecting the swinging of the leaves are based on the following two features of determining one of : (1) movement trajectory tracking, when the movement of the goal in path spot area of a region corresponding to the region of movement of the movement portion is less than the threshold value of the area of a region 5 time, the goal is that the swinging of the leaves; the target, for example, a 10 point track, these path spot only once in the sport is the region corresponding to the, as the swinging leaves this goal, the target filtering. (2) the amplitude of the movement of the mass center, if a certain target centroid of the amplitude of the movement is abrupt, the goal is that the swinging of the leaves, that is, when the adjacent path spot displacement of the mass center of the target in a target variation exceeds the threshold value of the width of the 6 time, that the leaves is that a swing of this goal, the target filtering.

The threshold value 5 is preferably 5%-15% between ; the threshold 6 is preferably 1.5-2.5 between.

Swing leaf shadow is detected by detecting the density of points in the region of to realize, the swinging leaves shadow detection method is: statistical expansion operation respectively before and after the number of points in a region (that is, the regional internal-expansion operating front and back pixel value of " 1 the number of the points of [...]), and calculating a ratio of them, if the ratio is less than the threshold value 7, that the shadow of this area is the area of the swinging leaves, and the filtering of the region.

The threshold value 7 is preferably 40%-60% between.

Figure 2 split and merged regions in the 16 is to use the background model provides the restraint of a priori knowledge of a man and vehicle model and the region such as a joint and splittist processing, in order to solve the target too apart and objectives on each other. The method of splitting the merged regions based on the above-mentioned reinforced area 15 treatment process, determining two adjacent regions of a target area is the same, or different target area. If it belongs to the same target area, then the two of the region merging; otherwise, its split. Wherein, two adjacent region is less than the threshold distance to the edge region 8 of the regional, with a regional index segni of coherent regional, different target area indicators the area of the mark is not consistent.

The threshold 8 preferably 3-7 between pixels.

2nd step of the present invention is to predict a target 20, the cumulative displacement according to the target movement and its corresponding accumulation time, calculating the average of the target moving speed, and on the basis of the speed prediction goal of the next displacement. Wherein, the accumulated displacement is the displacement of the of the target moving accumulated and, the accumulation time is of the target moving accumulated and of time. The accumulated displacement, and accumulating time in relationship to the average moving speed:

V=s/t

Wherein, the target s movement of the mass center of the stable displacement of the multiframe, the goal of the multiframe t the time required for the movement, for the v the average velocity of the stable target movement. Through the above-mentioned formula the average speed can be calculated.

Predicting v according to the average speed of the next displacement is:

S ' =v·Δt

Wherein, the target time to predict Δt, s ' Δ t the goal of stable movement of the mass center of the displacement of the time. Can be through the above-mentioned formula to calculate and predict the next displacement.

3rd step of the present invention that match the target 30, the stability of the matching used for tracking the target, and filtering the false target. Figure 4 is the invention in the process of matching goal schematic, as shown in Figure 4. Matching target 30 including: tracking matching the stable target 301, and to filter out false target 302.

Tracking matching the stable target 301 is to judge detection region is matched with the target tracking. The matching judgment conditions for: detecting region D the target matching coefficient of the calculation formula is as follows:

D=Da*A_Da +Db*A_Db +Dc*A_Dc

Wherein, an area matching coefficient Da, histogram matching coefficient for Db, Dc a distance matching coefficient. When the detecting region D the target matching coefficient is larger than the threshold 9, the, it is judged that the detection region is matched with the target. A_Da, A_Db, A_Dc are Da, Db, the weight coefficient corresponding to Dc. The threshold value 9 is preferably 0.7-0.8 between.

The stated A_Da, A_Db, A_Dc the value of 0-1 between, and meet the three and for the value of 1. The stated A_Da, A_Db, A_Dc the preferred value of 0.2, 0.3, 0.5.

1) area Da matching coefficient. When the detecting region of the target area that intersects the target area is greater than the threshold value of the area of the 10 time, the detection region that satisfies area matching, the Da 1 ; otherwise, the Da 0.

The threshold 10 is preferably 40%-60% between.

2) Db histogram matching coefficient. When the detecting region of the target area that intersects the histogram of the histogram is greater than the target threshold value of 11 when, that meet the detecting area matching of the histogram, taking Db 1 ; otherwise Db heating 0.

The threshold value 11 is preferably 40%-60% between.

3) distance match coefficient Dc. Sub-two situations to consider distance match coefficient Dc, the two conditions of the movement of the detection region is also is stationary. If the current frame in the previous frame of Image sensing area in the difference Image, the number of the former spots background spot number of threshold value is larger than 12, the, is that the movement of the detection region, otherwise that the detection area is stationary. When the detection area is the movement, calculate the current frame of Image detecting the center of the area to be detected with the current frame the distance of the center of the sensing area, if smaller than the target length of a diagonal line of the rectangular frame 13 of the threshold value, the distance that the matching of the meeting, taking Dc 1 ; otherwise Dc heating 0. When the detection area is stationary, the front one frame of the Image calculated detecting the center of the area to be detected with the current frame the distance of the center of the area to be examined, if it is less than the threshold value 14, the distance that the matching of the meeting, taking Dc 1 ; otherwise Dc heating 0.

The threshold 12 is preferably 65%-75% between. The threshold 13 is preferably 1.5-2 between. The threshold value 14 is preferably 8-12 between pixels.

Filtering the false target is that the track analysis of the target moving, in order to filter out false of the target area. Wherein, by using the target track is track analysis information (including a plane information and center-of-mass point information), statistical area change the smoothness and center-of-mass point of the stability of the change.

Wherein, the smoothness of the statistical the area change of the method is as follows: the area of statistical target track points set on {area₁, area₂, ... , area_n}, n that the number of the track points, mean value of statistical area:

Area [! OverBar! ] = 1 n &# X003A3; i = 1 n Area i

Statistical area variance: Area Sd = 1 n &# X003A3; i = 1 n ( Area i - Area [! OverBar! ] ) 2

When the area_sd/area> 0.5 time, that area does not change smoothly, the filtering the target area.

The stability of the statistical center-of-mass point changes in accordance with the normal objective of the method is in the direction of the movement of the will not produce recurrent mutation, in statistical adjacent path spot the ratio of the change of direction, if the ratio exceeds the threshold 15, the center-of-mass point is changed is not stable, the filtering the target area.

The threshold value 15 is preferably 40%-60% between.

The final step is to update the target 40, according to the target matching 30 the stability of the target, real-time updating of the model of the target tracking.

The invention also provides a moving target tracking system, Figure 5 is a schematic diagram of the structure of the target tracking system, as shown in Figure 5. Moving target tracking system comprises a detection target module 71, predict a target module 72, matching the target module 73 and updated target module 74. Wherein, the detection target module 71 is used for the movement of the video scene in the Image of the target area from the background, predict a target module 72 is used for estimating the motion of the target in the Image of the next frame the position of the scene, matching the target module 73 matching the stability of the used for tracking the target, and filtering the false target, updated target module 74 for updating the template stability objectives the current frame.

Figure 6 for this invention for detecting the moving target tracking system schematic diagram of the structure of the target module. As shown in Figure 6, the detection target module 71 includes obtaining the video module 711, pre-processes the Image module 712, mark zone module 713, maintenance state module 714, strengthening of the regional module 715 and splittist and integration of regional module 716. Wherein, to obtain video module 711, is used for obtaining the video content in order to get the scene Image, and the establishment of the background model; pre-processes the Image module 712, scene Image is used for eliminating the impact of the background model; mark zone module 713, according to the background model used for foreground segmentation of the scene Image, and mark the connected region; maintaining a state module 714, detecting target module for determining the state of the current, to make the corresponding processing, and the abnormality detection is necessary; enhancing regional module 715, used for using the shadow detection, high brightness detection and tree filter, eliminating shadow, high brightness and leaves false area of the; division and integration of regional module 716, is used for using the background model provides the restraint of the a priori knowledge of the people and vehicle model and the regional are combined and split and process, in order to solve the objectives and excessive segmentation on each other.

Figure 7 the movement of the target tracking system of the present invention in the structure of the matching the target module schematic diagram. As shown in Figure 7, the matching target module 73 comprises a tracking matching the stable target module 731 and filtering the false target module 732. Wherein, tracking matching the stable target module 731, is used for judging the detection region is matched with the target tracking, filtering the false target module 732, false region used for filtering.

The biggest advantage of the present invention under complex background lies in realizing the accurate tracking of a plurality of objects, solves the problem that the weatherstripping, leaves the question swinging, and operation is simple, it has very strong practicability.

The advantages of the invention also lies in the accurate detection of the moving object in the Image scene, include a person, vehicle, picture jitter can be neglected at the same time, the swing of the tree, the brightness variations, shadow, rain, snow, the impact of the interference factors.

The invention can also be used for intelligent video monitoring system, in order to achieve the target classification identification, alert moving target, moving target tracking, the tracking PTZ, automatic close-up shooting, target behavior detection, flow detection, congestion detection, carry-over detecting, is robbed the thing detection, smoke detection and flame detection functions.

Of the above, for this invention only the better practical example , and non-used for limiting the scope of protection of the present invention, it should be understood, the invention is not limited to the implementations described herein, the purpose of these implementations is described to the technical personnel in the field of practice of the invention. In any of the field of the technical staff easily without deviating from the spirit and scope of this invention to further improve and perfect, therefore, the invention only by the requirements of the invention and the scope of the rights of the content, the intention to cover all included in the defined by the attached claims and the spirit of the present invention within the range of options and equivalent of the programme.