CN110189365B - Anti-occlusion correlation filtering tracking method - Google Patents
Anti-occlusion correlation filtering tracking method Download PDFInfo
- Publication number
- CN110189365B CN110189365B CN201910438677.6A CN201910438677A CN110189365B CN 110189365 B CN110189365 B CN 110189365B CN 201910438677 A CN201910438677 A CN 201910438677A CN 110189365 B CN110189365 B CN 110189365B
- Authority
- CN
- China
- Prior art keywords
- target
- frame
- tracking
- occlusion
- proposal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/223—Analysis of motion using block-matching
- G06T7/238—Analysis of motion using block-matching using non-full search, e.g. three-step search
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
When video tracking is carried out, target classification is carried out according to a target response value and an area ratio of a second frame of a target to be tracked, tracking is stopped when whether the target response values of five adjacent frames meet an occlusion criterion or not, target detection is carried out on each frame of subsequently input images by starting a re-detection mechanism to obtain a plurality of proposal frames, and then the proposal frame with the maximum response value is obtained by preliminary screening and relevant filtering and is used for reinitializing a tracker so as to avoid detection misjudgment.
Description
Technical Field
The invention relates to a technology in the field of image processing, in particular to an anti-occlusion related filtering tracking method.
Background
Target tracking is one of the research hotspots in the field of computer vision. The method is widely applied to the fields of face recognition, behavior analysis, robotics, intelligent monitoring and the like. The existing target tracking algorithm has achieved great achievement, but due to factors such as posture change, illumination change, partial shielding, rapid movement, scale change and background complexity, many problems still exist in accurate target tracking.
In recent years, correlation filters have been introduced into the discriminant tracking framework and have achieved good results. The MOOSE (Minimum Output Sum of Squared Error) filter introduces correlation operation into target tracking, and greatly accelerates calculation through the theory that spatial domain convolution becomes the Hadamard product of Fourier domain. After that, the CSK (circular Structure of tracking-by-detection with Kernels) algorithm adopts a cyclic matrix to increase the number of samples, thereby improving the effect of the classifier. As an extension to CSK, oriented gradient features, gaussian kernel and ridge regression are used for kernel Correlation filter KCF (kernel Correlation Filters). Aiming at the scale change of the target, the problem of scale estimation is solved by identifying a scale space tracking (DSST) through a scale pyramid learning related filter. Long-term Correlation Tracking (LCT) includes appearance and motion Correlation filters to estimate scale and translation of objects. Inspired by human recognition models, choi proposed the ACFN algorithm (authorized feature-based Correlation Filter) to track rapidly changing targets. However, none of the above trackers solves the problem of target occlusion well or only partial occlusion (target occlusion area 50% or less of the total target area) and short-term full occlusion for the target. Moreover, the existing occlusion criterion cannot be well fused with a tracking algorithm, and the occlusion criterion can be judged wrongly in many times, which seriously influences the performance of the tracker.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides an Anti-occlusion related filtering tracking method, which realizes robust tracking under the situation that the target is occluded by the Anti-occlusion related filtering tracking (AO-CF), an occlusion standard based on a continuous response value and a new detection standard for detecting a potential reliable target.
The invention is realized by the following technical scheme:
the invention relates to an anti-occlusion correlation filtering tracking method, which is characterized in that during video tracking, target classification is carried out according to a target response value and an area ratio of a second frame of a target to be tracked, tracking is stopped when whether the target response values of five adjacent frames meet occlusion criteria or not, target detection is carried out on each frame of subsequently input images by starting a re-detection mechanism to obtain a plurality of proposal frames, and then the proposal frame with the maximum response value is obtained by primary screening and correlation filtering and used for reinitializing a tracker so as to avoid detection misjudgment.
The second frame of the target to be tracked refers to: and manually selecting a frame next to the frame where the target to be tracked is located.
The adjacent five frames refer to: the current frame of the target is taken as a starting point and then 4 continuous frames are added, so that an adjacent 5 frames are formed.
The target response valueWherein: y is the output of the sample, κ is the gaussian kernel, λ is the regularization coefficient, and the operator F represents the fourier transform.
The area ratio, i.e. the ratio of the area of the object to the whole pictureWherein: s. the 1 Is the area of the initial target frame, S 2 Is the area of the whole picture.
The target classification is as follows: after obtaining tau, mu, byPerforming a classification in which: n is a sign of and operation, a 1 And b 1 Respectively, threshold values used for classifying the targets; the coefficient d of the occlusion threshold can be further set according to the classified targets, where d = [ d ] 1 ,d 2 ,d 3 ,d 4 ]D mapping of elementsThe correlation between the beams is:
the shielding criterion is as follows: comparing the target response values of five continuous frames with an occlusion threshold, then counting the 5 response values to be less than the number, and finally judging whether the target is occluded or failed to track according to the comparison result and the statistical result, wherein the criterion is Y = [ Y (1), Y (2), Y (3), Y (4), Y (5)]D & tau, sum (Y & lttheta & d & tau) is more than or equal to 2, and theta & lt 1, wherein: y (i) is the response value, Y (i) is the element of Y, θ is the coefficient, and the operator sum (-) is used to calculate the number of Y (i) < θ d · τ in the set Y; coefficient d = [ d ] of occlusion threshold obtained according to target classification 1 ,d 2 ,d 3 ,d 4 ](ii) a The target is considered occluded when the response values of five consecutive frames reach the above condition.
The re-detection mechanism comprises the following specific steps:
1) When the target is judged to be shielded, an initial edge image of a next input frame image is obtained;
2) Combining adjacent edge pixels having the same direction to form a group;
3) Based on two edge groups s i ,s j Average position x of epsilon S i ,x j And average direction theta i ,θ j Calculating the similarity a(s) i ,s j ) The high similarity groups are connected into long contours according to the target classification principle: a(s) i ,s j )=|cos(θ i -θ ij )cos(θ j -θ ij )| γ Wherein: theta.theta. ij Is x i ,x j The angle between the two, gamma, is used to adjust the sensitivity of the change in direction.
The edge group is as follows: the edge group is composed of a group of edge points (pixels) with certain similarity.
The high similarity refers to that: root of herbaceous plantAccording to theta ij Is x i ,x j The similarity is judged according to the included angle between the two groups, and the smaller the included angle is, the higher the similarity is, and the smaller the similarity is.
4) The likelihood of the existence of the target is reflected by calculating the edge strengths of all the edge groups contained in the proposal box:wherein: w is a b (s i )∈[0,1]Is s is i Confidence in box b, m i Is in group s i Sum of the intensity of all edges, b w And b n To propose the width and height of box b, ζ is used to compensate for the deviation of a large window with more edges;
5) Subtracting the intensity of the set of edges that pass out of the detection box from the edge intensity, wherein: b in Is half the width and height of b;
6) Finding all detection frames by using a sliding window, and selecting a plurality of detection frames with higher scores to be arranged from high to low according to score values:
7) According to the position, the scale and the aspect ratio, iterative search is carried out by using a greedy principle, and the maximum value is obtainedThe meaning is to find another edge point with the smallest direction angle difference in the 8 neighborhoods of the edge points, then connect the two points, and then carry out the same operation on the other edge point, which can form the edge contour of the potential object and be used for scoring the confidence of the proposed box.
The preliminary screening is as follows: extracting a candidate target frame from the proposal frame by using a method for limiting the area, which specifically comprises the following steps: with the constraint conditions:to perform a preliminary screening of the proposal box, wherein: b is a mixture of w And b h Are respectivelyPropose the width and height of the frame->And &>Is the width and height of the target bounding box when the target is occluded.
The related filtering means: a relevant filtering operation is performed on each proposed box that passes the preliminary screening to obtain a corresponding response value.
The reinitialization is as follows: the maximum of the response values is compared to a threshold v τ and when the response value exceeds the threshold, the tracker is reinitialized with the center coordinates of its corresponding proposal box, thereby avoiding detection false positives.
In order to effectively detect the potential target, the selection condition of the threshold value comprises: tau is occ3 <τ occ2 <τ occ1 ,w<τ occ1 -τ occ3 ,Wherein: tau is occ1 ,τ occ2 And τ occ2 The method comprises the steps of respectively triggering response values of the first three frames after a standard, w represents the minimum difference value of the response values between the first frame and the third frame, r is used for measuring the descending degree of the response value of the second frame, the larger r is, the faster the response value of the second frame is, and then, different threshold coefficients are set to determine whether the detected boundary frame comprises a target desired by the method.
The invention relates to a system for realizing the method, which comprises the following steps: image input module, tracker module, re-detector module and output module, wherein: the image input module is connected with the tracker module and transmits the image and the position and width and height information of the target frame, the tracker module is connected with the re-detector module and transmits the image information, the tracker module is connected with the output module and transmits the image and the position and width and height information of the target frame, the re-detector module is connected with the output module and transmits the image and the position and width and height information of the target frame, and the re-detector module is connected with the tracker module and transmits the image and the position and width and height information of the target frame.
Drawings
FIG. 1 is a schematic diagram of a theoretical framework of an anti-occlusion target tracking method based on correlation filtering constructed by the present invention;
FIG. 2 is a graph illustrating a correlation filter response curve caused by 11 influencing factors according to the present invention;
in the figure: the sequence of (a) is sequence Shaking, (b) is sequence Toy, (c) is sequence Box, (d) is sequence Human5, (e) is sequence Jumping, (f) is sequence Boy, (g) is sequence Sylvester, (h) is sequence Lemming, (i) is sequence Couple, (j) is sequence Bird1, and (k) is sequence Panda;
FIG. 3 is a schematic diagram of the results of the primary screening for retesting in accordance with the present invention;
in the figure: (a) The proposal boxes left after the primary screening, (b) all the proposal boxes given by the re-detector;
FIG. 4 is a schematic diagram of a five-frame response curve of a general target when the occlusion criterion is triggered according to the present invention;
FIG. 5 is a schematic diagram of a five-frame response curve of a small target when the occlusion criterion is triggered according to the present invention;
FIG. 6 is a schematic diagram of the final determination result of the re-inspection according to the present invention;
FIG. 7 is a schematic diagram comparing the results of the present invention on OTB50 data set with other 9 robust tracking methods on tracking accuracy index;
FIG. 8 is a diagram illustrating comparison of results of the OTB50 data set and 9 other robust tracking methods on the overlay success rate index according to the present invention;
FIG. 9 is a schematic diagram comparing the results of the present invention on OTB100 data set with the results of other 9 robust tracking methods on tracking accuracy index;
FIG. 10 is a schematic diagram comparing the results of the invention on OTB100 data set with other 9 robust tracking methods on the overlap success rate index;
FIG. 11 is a schematic diagram comparing the tracking accuracy results under the illumination property of the OTB100 and other 9 robust tracking methods according to the present invention;
FIG. 12 is a schematic diagram comparing the tracking accuracy results of the invention on OTB100 with other 9 robust tracking methods under the out-of-plane rotation property;
FIG. 13 is a schematic diagram comparing the tracking accuracy results of the invention on OTB100 with other 9 robust tracking methods under the scale variation property;
FIG. 14 is a schematic diagram comparing the tracking accuracy results of the invention on OTB100 with other 9 robust tracking methods under the occlusion property;
FIG. 15 is a schematic diagram comparing the tracking accuracy results of the present invention on OTB100 with other 9 robust tracking methods under the distortion property;
FIG. 16 is a schematic diagram comparing the tracking accuracy results of the invention on OTB100 with other 9 robust tracking methods under the motion blur property;
FIG. 17 is a schematic diagram comparing the tracking accuracy results of the invention on OTB100 with other 9 robust tracking methods under the fast motion property;
FIG. 18 is a schematic diagram comparing the tracking accuracy results of the invention on OTB100 with other 9 robust tracking methods under the in-plane rotation property;
FIG. 19 is a schematic diagram comparing the tracking accuracy results of the present invention on OTB100 with other 9 robust tracking methods under out-of-view property;
FIG. 20 is a schematic diagram comparing the tracking accuracy results of the present invention on OTB100 with other 9 robust tracking methods under the background interference attribute;
FIG. 21 is a schematic diagram comparing the tracking accuracy results of the present invention on OTB100 with other 9 robust tracking methods under low resolution property;
FIG. 22 is a schematic diagram comparing the tracking overlap ratio results of the invention on OTB100 with other 9 robust tracking methods under the illumination property;
FIG. 23 is a schematic diagram comparing the tracking overlap ratio results of the invention on OTB100 under the out-of-plane rotation property with other 9 robust tracking methods;
FIG. 24 is a schematic diagram comparing the tracking overlap ratio results of the invention on OTB100 with other 9 robust tracking methods under the scale variation property;
FIG. 25 is a schematic diagram comparing the tracking overlap ratio results under the occlusion property of the OTB100 and other 9 robust tracking methods according to the present invention;
FIG. 26 is a schematic diagram comparing the tracking overlap ratio results of the invention on OTB100 with other 9 robust tracking methods under the distortion property;
FIG. 27 is a schematic diagram comparing the tracking overlap ratio results of the OTB100 and other 9 robust tracking methods under the motion blur property;
FIG. 28 is a schematic diagram comparing the tracking overlap ratio results of the invention on OTB100 with other 9 robust tracking methods under the fast motion property;
FIG. 29 is a schematic diagram comparing the tracking overlap ratio results of the invention on OTB100 under the in-plane rotation property with other 9 robust tracking methods;
FIG. 30 is a graph showing the comparison of tracking overlap rate results of the present invention on OTB100 with other 9 robust tracking methods under out-of-view property;
FIG. 31 is a schematic diagram comparing the tracking overlapping rate result of the present invention on OTB100 with other 9 robust tracking methods under the background interference attribute;
FIG. 32 is a schematic diagram comparing the tracking overlap ratio results of the invention on OTB100 with other 9 robust tracking methods under the low resolution property;
FIG. 33 is a comparison of the tracking results of the present invention and 9 other robust tracking methods on 12 video sequences with severe occlusion properties in the OTB100 data set.
Detailed Description
The embodiment comprises the following steps:
step 1) constructing a theoretical frame of an anti-occlusion target tracking method based on relevant filtering as shown in FIG. 1;
step 2) setting a 1 And b 1 As a basic condition for object classification, a 1 =0.6,b 1 =0.005。
Step 3) manually selecting a tracking target, training a motion correlation filter w on the target, and finding out the optimal scale beta of the target i Go to the next frame, λ =10 in this embodiment -4 Gaussian kernel width σ =0.1;
and 4) recording the second frame response value tau of the target and the area ratio mu, and classifying the target.
TABLE 1
The relationships in the table corresponding to fig. 2 are: the sequence of the gene is (a) sequence Shaking, (b) sequence Toy, (c) sequence Box, (d) sequence Human5, (e) sequence Jumping, (f) sequence Boy, (g) sequence Sylvester, (h) sequence Lemming, (i) sequence Couple, (j) sequence Bird1, and (k) sequence Panda.
Step 5) setting corresponding occlusion coefficients according to different targets and according to classification thresholds, d 1 =0.3,d 2 =0.5,d 3 =0.4,d 4 =0.6。
And 6) establishing a new frame of target search area, wherein the position of the new frame of target search area is the same as that of the target frame of the previous frame, and the area of the new frame of target search area is 1.5 times that of the target frame.
And 7) extracting a characteristic vector x of the target, weighting by a cosine window, and jointly using the response graph and the target scale to simultaneously obtain the maximum response value under the motion and scale. Choose to have the largestAs a result of the translation estimation of the target. Meanwhile, selecting beta corresponding to the maximum response value i As the optimal scale for the target.
Step 8) judging whether the latest continuous five frames meet the occlusion criterion of the embodiment, namely:
1)Y=[y(1),y(2),y(3),y(4),y(5)]<d·τ
2)sum(Y<θ·d·τ)≥2,θ<1
in the present example, d = [ d ] 1 ,d 2 ,d 3 ,d 4 ],θ=0.7。
Step 9) when the target is determined to be shielded, detecting each frame of image input next by using Edge Boxes, and selecting the first k =200 detection frames with high score values as candidate frames for target detection in the embodiment. And filtering most unreasonable detection frames by using the constraint conditions of the candidate target frames. The filtered target box is shown in fig. 3.
Step 10) in the detection threshold coefficient setting, w =0.05,z 1 =z 2 =0.6. Setting coefficients v, v of detection threshold according to selection conditions of potential targets 1 =0.7,v 2 =0.5,v 3 =0.8,v 4 =0.6。
Step 11) for the reserved detection frames, using a response graph and a target scale to sequentially perform related filtering on the detection frames, and if the final response value of the highest one is greater than v tau, outputting the target frame as a new initial condition to start a tracker; otherwise, the next frame is entered for detection until the target is detected. The final test results are shown in fig. 6. And after the target frame of the current frame is obtained, entering the next frame, and repeating the steps until the image sequence is finished.
To verify the validity of this embodiment, this embodiment compares the proposed algorithm with the other 9 advanced trackers. These 9 advanced trackers are: KCF, IKCF, DSST, LCT, MEEM, SAMF, DLSSVM, stacke and ACFN, wherein: IKCF is the embodiment that the proposed occlusion criterion and re-detection mechanism are added into KCF to further prove the effectiveness of the proposed method. The experimental environment was Intel Core i 5.3GHz CPU with 8.00G RAM, MATLAB 2017b.
To evaluate the overall performance of the tracker, the present embodiment evaluates it on the disclosed target tracking benchmark (OTB) data set.
The target tracking reference OTB data set includes:
1) An OTB-50 having a sequence of 50,
2) OTB-100 with 100 sequences.
All these sequences are annotated with 11 attributes, covering various challenging factors including scale changes, occlusion, illumination changes, motion blur, morphing, fast motion, out-of-plane rotation, background clutter interference, out-of-view, in-plane rotation, and low resolution. The present embodiment uses two indices in the reference data set to evaluate tracking performance, namely, the overlap success rate and the distance accuracy rate.
In fig. 8 and 9, it can be seen that the tracker of the present embodiment ranks the distance accuracy rate at the second name and ranks the distance accuracy rate at the first name at the overlapping success rate on the OTB-50 data set; in fig. 9 and 10, the tracker of this embodiment ranks the first two indicators on the OTB-100 data set. This fully demonstrates the effectiveness of the algorithm proposed by the present embodiment.
For 11 kinds of challenge attributes, as shown in fig. 11 to fig. 21, on the distance accuracy index, the present embodiment ranks the first challenge attribute among seven challenge attributes of illumination, out-of-plane rotation, scale change, occlusion, in-plane rotation, background interference, and low resolution, and ranks the second challenge attribute below the distorted challenge attribute.
For 11 kinds of challenge attributes, as shown in fig. 22 to fig. 32, on the index of the overlapping success rate, the present embodiment ranks the first challenge attribute among six challenge attributes of out-of-plane rotation, occlusion, in-plane rotation, out-of-view, background interference, and low resolution, and ranks the second challenge attribute among four challenge attributes of illumination, scale change, distortion, and background blur. Therefore, the problem of target shielding is well solved, and meanwhile the problem of tracking drift caused by other factors is effectively solved.
As shown in tables 2 to 4, the tracker of the present embodiment is most excellent in terms of the overall tracking performance and the tracking performance under each attribute. On the OTB-100 data set, the accuracy index of the tracker of the embodiment reaches 81%, and the index of the overlapping success rate reaches 58.9%. In addition, the accuracy index of the embodiment exceeds 1.3 percent of that of the second ACFN tracker, and the overlapping success rate index of the embodiment exceeds 0.8 percent of that of the second Stacke tracker. In particular, under the occlusion property, the tracker of the present embodiment exceeds the second ACFN tracker by 3.4 percentage points in the accuracy index, and exceeds the second Staple tracker by 2.3 percentage points in the overlapping success rate.
TABLE 2
TABLE 3
TABLE 4
OCC, IV, DEF, OPR, BC, SV, MB, LR, FM in the tables have the following specific meanings: occlusion attribute, illumination variation attribute, distortion attribute, out-of-plane rotation attribute, background interference attribute, scale variation attribute, motion blur attribute, low resolution attribute, fast motion attribute.
In qualitative assessment for occlusion property, this embodiment provides that 12 video sequences with severe occlusion property in the OTB100 data set are selected, and the basic information of the sequences is shown in fig. 33.
As shown in FIG. 33, the AO-CF is still tracking the target robustly when it is undergoing partial occlusion or full occlusion. However, most trackers drift into the background after the target is occluded. For IKCF, the improved algorithm enables robust tracking of eight sequences. IKCF achieves a significant performance boost compared to KCF. For the 17 th frame of the basketball sequence, SAME, IKCF and DSST drift after the target is occluded. But with cooperation of the re-detectors, the IKCF can quickly retrieve the target and resume correct tracking. For the Box sequence, it is clear that only SAME, IKCF, and AO-CF are able to achieve accurate tracking, while other trackers experience severe drift. For frame 131 of the Coupon sequence, when the target is occluded and a similar target appears, MEEM and DLSSVM lose the target while the other trackers work well. In the sequences Walking2, human4, lemming and Panda, the target has a significant change in appearance in addition to OCC. Since IKCF has no scale-adaptive function, the template is gradually contaminated with background information, and satisfactory tracking results were obtained for the sequences Jogging-1, kitesurf and Tiger2, AO-CF and IKCF, where: the targets in the Kitesurf and Tiger2 sequences also suffered from severe illumination changes, which also suggests that the proposed algorithm is well adapted to illumination changes. For the Girl2 sequence, when the target is completely occluded, only the AO-CF and the IKCF accurately restore the target, and all other trackers cannot perform normal tracking. For the most challenging sequences of soccer balls, the target is disturbed by multiple attributes, and even in such a complex environment, AO-CF and IKCF can still track the target normally.
The foregoing embodiments may be modified in many different ways by one skilled in the art without departing from the spirit and scope of the invention, which is defined by the appended claims and not by the preceding embodiments, and all embodiments within their scope are intended to be limited by the scope of the invention.
Claims (7)
1. An anti-occlusion correlation filtering tracking method is characterized in that during video tracking, target classification is carried out according to a target response value and an area ratio of a second frame of a target to be tracked, tracking is stopped when target response values of five adjacent frames meet occlusion criteria, a re-detection mechanism is started to carry out target detection on each frame of subsequently input images and obtain a plurality of proposal frames, and then the proposal frame with the largest response value is obtained through preliminary screening and correlation filtering and used for reinitializing a tracker so as to avoid detection misjudgment;
the target response valueWherein: y is the output of the sampling sample, k is a Gaussian kernel function, λ is a regularization coefficient, and an operator F represents Fourier transform;
the area ratio, i.e. the ratio of the area of the target to the whole pictureWherein: s. the 1 Is the area of the initial target frame, S 2 The area of the whole picture;
the re-detection mechanism comprises the following specific steps:
1) When the target is judged to be shielded, the initial edge image of the next input frame image is processed;
2) Combining adjacent edge pixels having the same direction to form a group;
3) Based on two edge groups s i ,s j Average position x of e S i ,x j And average direction θ i ,θ j Calculating the similarity a(s) i ,s j ) The high similarity groups are connected into long contours according to the target classification principle: a(s) i ,s j )=|cos(θ i -θ ij )cos(θ j -θ ij )| γ Wherein: theta ij Is x i ,x j The included angle between the two, gamma is used for adjusting the sensitivity of the direction change;
4) The likelihood of the existence of the target is reflected by calculating the edge strengths of all the edge groups contained in the proposal box:wherein: w is a b (s i )∈[0,1]Is s is i Confidence in box b, m i Is in group s i Sum of the intensity of all edges, b w And b h To propose the width and height of box b, ζ is used to compensate for the deviation of a large window with more edges;
5) Subtracting the intensity of the set of edges that pass out of the detection box from the edge intensity, b in Is half the width and height of b;
6) Finding all detection frames by using a sliding window, and selecting a plurality of detection frames with higher scores to be arranged from high to low according to score values:
7) According to the position, the scale and the aspect ratio, iterative search is carried out by using a greedy principle to obtain the maximumThe meaning is that another edge point with the smallest direction angle difference is found in the neighborhood of 8 of one edge point, then the two points are connected, and then the same operation is carried out on the other edge point, and the operation can form the edge contour of the potential object and is used for scoring the confidence degree of the proposal box;
the shielding criterion is as follows: comparing the target response values of five continuous frames with an occlusion threshold, then counting the 5 response values to be less than the number, and finally judging whether the target is occluded or failed to track according to the comparison result and the statistical result, wherein the criterion is Y = [ Y (1), Y (2), Y (3), Y (4), Y (5)]D & tau, sum (Y & lttheta & d & tau) is more than or equal to 2, and theta & lt 1, wherein: y (i) is the response value, Y (i) is the element of Y, θ is the coefficient, and the operator sum (-) is used to calculate the number of Y (i) < θ · d · τ in the set Y; coefficient d = [ d ] of occlusion threshold obtained according to target classification 1 ,d 2 ,d 3 ,d 4 ](ii) a The object is considered to be occluded when the response values of five consecutive frames reach the above condition.
2. The method of claim 1, wherein said object classification is: after obtaining tau, mu, byPerforming classification, wherein: n is a sign of and operation, a 1 And b 1 Respectively, threshold values used to classify the target; the shielding can be further arranged according to the classified targetsCoefficient of gear threshold value d, wherein d = [ d = [ d = 1 ,d 2 ,d 3 ,d 4 ]And the mapping relation of the elements in the d is as follows: />d 1 <d 2 ,d 3 <d 4 ,d 1 <d 3 ,d 2 <d 4 。
3. The method of claim 1, wherein said preliminary screening is: extracting a candidate target frame from the proposal frame by using a method for limiting the area, which specifically comprises the following steps: with constraint conditionsPerforming a preliminary screening of the proposal box, wherein: b w And b h Respectively, the width and height of the proposal frame>And &>Is the width and height of the target bounding box when the target is occluded.
4. The method of claim 1, wherein said correlation filtering is: and carrying out related filtering operation on each proposal box passing the preliminary screening to obtain corresponding response values.
5. The method of claim 1, wherein the re-initialization is: the maximum of the response values is compared to a threshold and when the threshold is exceeded, the tracker is reinitialized with the center coordinates of its corresponding proposal box, thereby avoiding detection false positives.
6. The method of claim 5, wherein the threshold is selected by the following conditions:
τ occ3 <τ occ2 <τ occ1 ,w<τ occ1 -τ occ3 ,wherein: tau is occ1 ,t occ2 And t occ2 The response values of the first three frames after the trigger criteria, w represents the minimum difference in response values between the first frame and the third frame, and r is used to measure the degree of decline in response values of the second frame.
7. A system for implementing the anti-occlusion related filtering tracking method of any one of claims 1-6, comprising: image input module, tracker module, re-detector module and output module, wherein: the image input module is connected with the tracker module and transmits the image and the position and width and height information of the target frame, the tracker module is connected with the re-detector module and transmits the image information, the tracker module is connected with the output module and transmits the image and the position and width and height information of the target frame, the re-detector module is connected with the output module and transmits the image and the position and width and height information of the target frame, and the re-detector module is connected with the tracker module and transmits the image and the position and width and height information of the target frame.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910438677.6A CN110189365B (en) | 2019-05-24 | 2019-05-24 | Anti-occlusion correlation filtering tracking method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910438677.6A CN110189365B (en) | 2019-05-24 | 2019-05-24 | Anti-occlusion correlation filtering tracking method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110189365A CN110189365A (en) | 2019-08-30 |
CN110189365B true CN110189365B (en) | 2023-04-07 |
Family
ID=67717759
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910438677.6A Active CN110189365B (en) | 2019-05-24 | 2019-05-24 | Anti-occlusion correlation filtering tracking method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110189365B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112489090B (en) * | 2020-12-16 | 2024-06-04 | 影石创新科技股份有限公司 | Method for tracking target, computer readable storage medium and computer device |
CN112950687B (en) * | 2021-05-17 | 2021-08-10 | 创新奇智(成都)科技有限公司 | Method and device for determining tracking state, storage medium and electronic equipment |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103077539A (en) * | 2013-01-23 | 2013-05-01 | 上海交通大学 | Moving object tracking method under complicated background and sheltering condition |
WO2016131300A1 (en) * | 2015-07-22 | 2016-08-25 | 中兴通讯股份有限公司 | Adaptive cross-camera cross-target tracking method and system |
CN106097393A (en) * | 2016-06-17 | 2016-11-09 | 浙江工业大学 | A kind of based on multiple dimensioned and adaptive updates method for tracking target |
CN106557774A (en) * | 2015-09-29 | 2017-04-05 | 南京信息工程大学 | The method for real time tracking of multichannel core correlation filtering |
CN107481264A (en) * | 2017-08-11 | 2017-12-15 | 江南大学 | A kind of video target tracking method of adaptive scale |
CN108198209A (en) * | 2017-12-22 | 2018-06-22 | 天津理工大学 | It is blocking and dimensional variation pedestrian tracking algorithm |
CN108665481A (en) * | 2018-03-27 | 2018-10-16 | 西安电子科技大学 | Multilayer depth characteristic fusion it is adaptive resist block infrared object tracking method |
CN108765470A (en) * | 2018-06-17 | 2018-11-06 | 天津理工大学 | One kind being directed to the improved KCF track algorithms of target occlusion |
CN108765452A (en) * | 2018-05-11 | 2018-11-06 | 西安天和防务技术股份有限公司 | A kind of detection of mobile target in complex background and tracking |
CN109102522A (en) * | 2018-07-13 | 2018-12-28 | 北京航空航天大学 | A kind of method for tracking target and device |
CN109299735A (en) * | 2018-09-14 | 2019-02-01 | 上海交通大学 | Anti-shelter target tracking based on correlation filtering |
CN109598684A (en) * | 2018-11-21 | 2019-04-09 | 华南理工大学 | In conjunction with the correlation filtering tracking of twin network |
-
2019
- 2019-05-24 CN CN201910438677.6A patent/CN110189365B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103077539A (en) * | 2013-01-23 | 2013-05-01 | 上海交通大学 | Moving object tracking method under complicated background and sheltering condition |
WO2016131300A1 (en) * | 2015-07-22 | 2016-08-25 | 中兴通讯股份有限公司 | Adaptive cross-camera cross-target tracking method and system |
CN106557774A (en) * | 2015-09-29 | 2017-04-05 | 南京信息工程大学 | The method for real time tracking of multichannel core correlation filtering |
CN106097393A (en) * | 2016-06-17 | 2016-11-09 | 浙江工业大学 | A kind of based on multiple dimensioned and adaptive updates method for tracking target |
CN107481264A (en) * | 2017-08-11 | 2017-12-15 | 江南大学 | A kind of video target tracking method of adaptive scale |
CN108198209A (en) * | 2017-12-22 | 2018-06-22 | 天津理工大学 | It is blocking and dimensional variation pedestrian tracking algorithm |
CN108665481A (en) * | 2018-03-27 | 2018-10-16 | 西安电子科技大学 | Multilayer depth characteristic fusion it is adaptive resist block infrared object tracking method |
CN108765452A (en) * | 2018-05-11 | 2018-11-06 | 西安天和防务技术股份有限公司 | A kind of detection of mobile target in complex background and tracking |
CN108765470A (en) * | 2018-06-17 | 2018-11-06 | 天津理工大学 | One kind being directed to the improved KCF track algorithms of target occlusion |
CN109102522A (en) * | 2018-07-13 | 2018-12-28 | 北京航空航天大学 | A kind of method for tracking target and device |
CN109299735A (en) * | 2018-09-14 | 2019-02-01 | 上海交通大学 | Anti-shelter target tracking based on correlation filtering |
CN109598684A (en) * | 2018-11-21 | 2019-04-09 | 华南理工大学 | In conjunction with the correlation filtering tracking of twin network |
Non-Patent Citations (5)
Title |
---|
An Anti-occlusion Correlation Filtering Tracking Algorithm for UAV;Zun Xu等;《IEEE》;20190506;全文 * |
基于核相关滤波的视觉跟踪算法研究;顾培婷;《中国优秀硕士学位论文全文数据库信息科技辑》;20180115(第1期);全文 * |
基于相关滤波的视频目标跟踪技术研究;王艳川;《中国优秀硕士学位论文全文数据库信息科技辑》;20181215(第12期);全文 * |
抗遮挡核相关滤波目标跟踪算法研究;张杨;《中国优秀硕士学位论文全文数据库信息科技辑》;20181215(第12期);全文 * |
改进的KCF目标跟踪算法研究;刘思思;《中国优秀硕士学位论文全文数据库信息科技辑》;20190415(第4期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN110189365A (en) | 2019-08-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111460926B (en) | Video pedestrian detection method fusing multi-target tracking clues | |
CN110599519B (en) | Anti-occlusion related filtering tracking method based on domain search strategy | |
Hua et al. | Online object tracking with proposal selection | |
Zhou et al. | Robust vehicle detection in aerial images using bag-of-words and orientation aware scanning | |
Fernandez-Sanjurjo et al. | Real-time visual detection and tracking system for traffic monitoring | |
Gao et al. | A deep learning based approach to classification of CT brain images | |
CN101981582B (en) | Method and apparatus for detecting object | |
CN111582349B (en) | Improved target tracking algorithm based on YOLOv3 and kernel correlation filtering | |
CN105893946A (en) | Front face image detection method | |
CN105893957B (en) | View-based access control model lake surface ship detection recognition and tracking method | |
CN110033472B (en) | Stable target tracking method in complex infrared ground environment | |
CN109191488B (en) | Target tracking system and method based on CSK and TLD fusion algorithm | |
CN110189365B (en) | Anti-occlusion correlation filtering tracking method | |
CN106570490A (en) | Pedestrian real-time tracking method based on fast clustering | |
Bouachir et al. | Structure-aware keypoint tracking for partial occlusion handling | |
CN111027445B (en) | Marine ship target identification method | |
WO2013075295A1 (en) | Clothing identification method and system for low-resolution video | |
CN108898621B (en) | Related filtering tracking method based on instance perception target suggestion window | |
CN108460336A (en) | A kind of pedestrian detection method based on deep learning | |
CN113850221A (en) | Attitude tracking method based on key point screening | |
JP7392488B2 (en) | Recognition method, device, and image processing device for false detection of remains | |
Liu et al. | Correlation filter with motion detection for robust tracking of shape-deformed targets | |
Wu et al. | Video smoke separation and detection via sparse representation | |
CN116109950A (en) | Low-airspace anti-unmanned aerial vehicle visual detection, identification and tracking method | |
CN116469020A (en) | Unmanned aerial vehicle image target detection method based on multiscale and Gaussian Wasserstein distance |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |