CN108268823B - Target re-identification method and device - Google Patents
Target re-identification method and device Download PDFInfo
- Publication number
- CN108268823B CN108268823B CN201611265859.0A CN201611265859A CN108268823B CN 108268823 B CN108268823 B CN 108268823B CN 201611265859 A CN201611265859 A CN 201611265859A CN 108268823 B CN108268823 B CN 108268823B
- Authority
- CN
- China
- Prior art keywords
- tracking
- image
- model
- information
- feature
- 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
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000012545 processing Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- HNYOPLTXPVRDBG-UHFFFAOYSA-N barbituric acid Chemical compound O=C1CC(=O)NC(=O)N1 HNYOPLTXPVRDBG-UHFFFAOYSA-N 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 abstract description 14
- 230000008569 process Effects 0.000 description 18
- 230000008859 change Effects 0.000 description 17
- 238000005286 illumination Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000005034 decoration Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000001174 ascending effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a target re-identification method and a target re-identification device. Wherein, the method comprises the following steps: acquiring a tracking target and an image area of the tracking target; extracting characteristic information from an image area of a tracking target, and constructing a characteristic model according to the characteristic information; determining the tracking state of a tracking target according to the reliability of the tracking result of the current frame image, wherein the reliability of the tracking result is determined by the similarity between the feature information of the preset area of the current frame image and the feature model; and under the condition that the tracking target is determined not to be lost according to the tracking state, updating the characteristic model according to the tracking result of the current frame image. The invention solves the technical problem of poor robustness of the re-identification technology of the tracked target in the existing tracking technology.
Description
Technical Field
The invention relates to the field of video image processing, in particular to a target re-identification method and device.
Background
Image-based object re-recognition generally refers to recognizing a given object from different images and videos, and the technology is generally used in the fields of object tracking across scenes, content-based image retrieval and the like. The conventional method is to use paired image data of the same target in different scenes and paired image data of different targets to respectively extract specified features, such as color histograms, as feature vectors, then learn a similarity measurement function by using a metric learning method, and in application, calculate the similarity of the two targets by using the similarity measurement function, so as to judge whether the two targets are the same target.
Recently, with the rise of deep learning, the appearance of target re-identification by using a convolutional neural network is also utilized, the idea is similar to that of the traditional method, the difference is that characteristics do not need to be specified artificially, the characteristics representing the similarity and difference of the target and the function for measuring the similarity of the characteristics are obtained by automatically learning by the convolutional neural network, and in the application, the learned convolutional network model is applied to two images so as to judge whether the target is the same.
The essence of visual tracking is to find the position of the same target between different frames, and generally speaking, a tracking-by-detection tracking system can be regarded as a target re-identification process when judging whether the targets are the same. However, no matter the traditional target re-identification method or the deep learning-based method, a similarity measurement function is obtained off line, and in application, whether two images are the same or not is directly judged. Due to the influence of the appearance change of the tracked target caused by the environment and illumination change in the tracking process, if the target re-identification mode is directly applied to a tracking system, whether the target is the same or not is judged by using two pictures, and the tracking system is often limited by the change of the environment; in addition, the problem of target re-identification in visual tracking is different from pure target re-identification to a certain extent, and the target re-identification in tracking needs to judge whether the target in a subsequent video frame is the same as that in initial setting, rather than finding the same target from an open set in a broad sense.
Generally, visual tracking maintains an online updating template, and finds a tracking target in a new frame by using the template, in the long-time tracking process, the method is influenced by the appearance change of the tracking target caused by the change of environment and illumination, tracking errors occur, the tracking error is amplified continuously and is difficult to correct, and one disadvantage of the method is that whether the tracking target is lost or not is difficult to accurately judge, or the initial tracking target is difficult to find back after the target is lost; in addition, after the target is lost, due to changes of illumination, environment and the like in the tracking process, the appearance of the target between frames is obviously changed, and the target is difficult to accurately find back through the appearance.
Aiming at the problem of poor robustness of the re-identification technology of the tracking target in the existing tracking technology, an effective solution is not provided at present.
Disclosure of Invention
The embodiment of the invention provides a target re-identification method and a target re-identification device, which at least solve the technical problem of poor robustness of the re-identification technology of a tracked target in the existing tracking technology.
According to an aspect of an embodiment of the present invention, there is provided a target re-identification method, including: acquiring a tracking target and an image area of the tracking target; extracting characteristic information from an image area of a tracking target, and constructing a characteristic model according to the characteristic information; determining the tracking state of a tracking target according to the reliability of the tracking result of the current frame image, wherein the reliability of the tracking result is determined by the similarity between the feature information of the preset area of the current frame image and the feature model; and under the condition that the tracking target is determined not to be lost according to the tracking state, updating the characteristic model according to the tracking result of the current frame image.
Further, original feature information in a preset model is replaced by feature information extracted from an image area of the tracking target, and a feature model is obtained.
Further, performing feature extraction on an image area of the obtained tracking result of the current frame image, and performing normalization processing to obtain a plurality of corresponding feature information; acquiring a preset probability; and replacing any one feature information in the feature model with a preset probability through preset feature information in the feature information corresponding to the obtained current frame image so as to update the feature model.
Further, acquiring the Bhattacharyya distance between the feature information of the current frame image and the feature model which is updated at the last time; obtaining a preset probability through the following formula:
wherein p is a predetermined probability, dmedianσ is a preset constant, which is the babbitt distance between the feature information and the feature model updated most recently.
Further, the median value of the plurality of babbitt distances between the feature information of the current frame image and the plurality of feature information in the latest updated model is determined as the babbitt distance between the feature information and the latest updated model.
Further, removing a background image in an image area of the tracking target; dividing the image area without the background image into a plurality of images along a preset direction; acquiring the characteristic information of the plurality of equally divided images; and connecting the feature information of the plurality of images after the average division according to the dividing sequence to obtain the feature information of the image of the tracking target.
Further, the feature information is image color feature information, wherein the image color feature information includes: color name information and/or hue information.
According to another aspect of the embodiments of the present invention, there is also provided an object re-recognition apparatus including: the acquisition module is used for acquiring a tracking target and an image area of the tracking target; the construction module is used for extracting characteristic information from the image area of the tracking target and constructing a characteristic model according to the characteristic information; the determining module is used for determining the tracking state of the tracking target according to the reliability of the tracking result of the current frame image, wherein the reliability of the tracking result is determined by the similarity between the feature information of the preset area of the current frame image and the feature model; and the updating module is used for updating the characteristic model according to the tracking result of the current frame image under the condition that the tracking target is determined not to be lost according to the tracking state.
Further, the construction module includes: and the initialization submodule is used for replacing original characteristic information in a preset model by the characteristic information extracted from the image area of the tracking target to obtain a characteristic model.
Further, the update module includes: the extraction submodule is used for extracting the characteristics of an image area for acquiring the tracking result of the current frame image and carrying out normalization processing to obtain corresponding characteristic information; the first obtaining submodule is used for obtaining a preset probability; and the replacing submodule is used for replacing any one piece of characteristic information in the characteristic model with a preset probability through the preset characteristic information in the acquired characteristic information corresponding to the current frame image so as to update the characteristic model.
Further, the first obtaining sub-module includes: the acquisition unit is used for acquiring the Pasteur distance between the feature information of the current frame image and the feature model which is updated last time; the calculating unit is used for acquiring the preset probability through the following formula:
wherein p is a predetermined probability, dmedianσ is a preset constant, which is the babbitt distance between the feature information and the feature model updated most recently.
Further, the acquisition unit includes: and the determining subunit is used for determining the median value of the plurality of the babbit distances between the feature information of the current frame image and the plurality of feature information in the latest updated model as the babbit distance between the feature information and the latest updated model.
Further, the construction module includes: the background removing submodule is used for removing a background image in an image area of the tracking target; the segmentation submodule is used for segmenting the image area without the background image into a plurality of images along a preset direction; the second obtaining submodule is used for obtaining the characteristic information of the plurality of equally divided images; and the connecting sub-module is used for connecting the feature information of the plurality of images after the average division according to the dividing sequence to obtain the feature information of the image of the tracking target.
Further, the feature information is image color feature information, wherein the image color feature information includes: color name information and/or hue information.
In the embodiment of the invention, a tracking target and an image area of the tracking target are obtained, feature information is extracted from the image area of the tracking target, a feature model is constructed according to the feature information, the tracking state of the tracking target is determined according to the reliability of the tracking result of the current frame image, and the feature model is updated according to the tracking result of the current frame image under the condition that the tracking target is determined not to be lost according to the tracking state. According to the scheme, the characteristic model is constructed according to the characteristic information of the image area of the opera tracking target, the characteristic model is continuously updated according to different tracking results in the tracking process, and the characteristic model is used as the tracking model for tracking, so that the robustness of the tracking model is improved, and the technical problem of poor robustness of the re-identification technology of the tracking target in the existing tracking technology is solved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:
FIG. 1 is a flow chart of a target re-identification method according to an embodiment of the invention; and
fig. 2 is a schematic diagram of an object re-recognition apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Example 1
In accordance with an embodiment of the present invention, there is provided an embodiment of an object re-identification method, it should be noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system such as a set of computer executable instructions, and that while a logical order is illustrated in the flowchart, in some cases the steps illustrated or described may be performed in an order different than that herein.
Fig. 1 is a flowchart of an object re-identification method according to an embodiment of the present invention, as shown in fig. 1, the method includes the following steps:
step S102, a tracking target and an image area of the tracking target are obtained.
Specifically, the tracking target may be a target specified by an operator or determined by a pedestrian detector, and the image area of the tracking target may be an area containing the tracking target indicated in an image of a certain frame of the video by an operator or an image area determined by a pedestrian detector in a certain frame of the video.
And step S104, extracting characteristic information from the image area of the tracking target, and constructing a characteristic model according to the characteristic information.
Specifically, the extracted features may be color features, edge features, and the like of an image, and since a tracking target is usually dynamic in a video, tracking only using the shape of the tracking target as a model has certain difficulty and is low in accuracy, but generally for continuous images in the video, the tracking target changes in time and the shape changes with the change of a timestamp, but the features of the images are usually consistent, and therefore the above steps construct the model by using the extracted image features.
And S106, determining the tracking state of the tracking target according to the reliability of the tracking result of the current frame image, wherein the reliability of the tracking result is determined by the similarity between the feature information of the preset area of the current frame image and the feature model.
Specifically, the tracking result includes a region and a reliability of the tracking target in the image, and the tracking state of the tracking target may include three states of non-loss, low reliability, and loss. In an alternative embodiment, a confidence threshold may be set, and if the confidence level of the tracking result is determined to exceed the preset confidence threshold, the tracking result is determined not to be lost. When the reliability of the tracking result is determined according to the similarity between the feature information of the preset region of the current image and the feature model, the reliability can be determined by using the first feature information of the image, or by using a plurality of feature information of the image after information fusion.
And step S108, under the condition that the tracking target is determined not to be lost according to the tracking state, updating the characteristic model according to the tracking result of the current frame image.
In an optional embodiment, the above scheme may be used in a process of tracking and determining loss, that is, a process of determining whether a tracking target is lost, in an optional embodiment, taking the tracking target as a specific person as an example, the tracking task is to track the specific person in a plurality of videos, after analyzing each frame image of the videos, obtaining a determination result of determining whether to follow loss, if it is detected that the similarity between the feature information of the preset region of the current frame image and the feature model is higher than a preset value, determining that the tracking is not followed loss, updating the feature model with the tracking result in the current frame image, and continuing to track, if it is detected that the similarity between the feature information of the preset region of the current frame image and the feature model is lower than the preset value, determining that the current frame image has followed loss, and entering a step of finding an original tracking target.
In another optional embodiment, the above scheme may be used in a process of retrieving an original tracking target after a tracking target is lost, for example, when it is determined that a current frame image is lost, a tracking target may be retrieved based on a tracking model that is updated recently, that is, a previous frame tracking model, and since the previous frame image is in a state of not being lost, a feature model of the tracking target is updated after a tracking result of the previous frame image is obtained, so that the feature model used for retrieving the tracking target is a closest feature model.
It should be noted here that, because the specific task may be in a dynamic state in the video, and other environmental information in the video also changes over time, that is, the shape of the specific task changes constantly in the video, and the illumination and environment in the video also change, it is very difficult to perform tracking or re-finding simply by tracking the shape of the target, and further, an accurate result cannot be obtained by performing tracking by always using the initially determined feature model of the tracked target, and therefore, the feature model of the tracked target introduced in the above scheme can effectively remove the influence of the change in the environment or the change in the shape of the tracked target in the tracking or re-finding process, thereby upgrading the robustness of the tracking model.
As can be seen from the above, in the present application, the tracking target and the image area of the tracking target are obtained in the above steps, the feature information is extracted from the image area of the tracking target, the feature model is constructed according to the feature information, the tracking state of the tracking target is determined according to the reliability of the tracking result of the current frame image, and the feature model is updated according to the tracking result of the current frame image when it is determined that the tracking target is not lost according to the tracking state. According to the technical scheme, the characteristic model is constructed according to the characteristic information of the image area of the tracking target, the characteristic model is continuously updated according to different tracking results in the tracking process, and the characteristic model is used as the tracking model for tracking, so that the robustness of the tracking model is improved, and the technical problem that the re-identification technology of the tracking target in the existing tracking technology is poor in robustness is solved.
Optionally, according to the above embodiment of the present application, in step S102, a feature model is constructed according to the feature information, including:
and step S1021, replacing original feature information in a preset model by feature information extracted from the image area of the tracking target to obtain a feature model.
In an optional embodiment, taking the feature information as color feature information of an image as an example, the corresponding feature model is a color feature model, a tracking target may be selected, color features are extracted from a selected target image region, where a color histogram is used as the color feature information, and an original model is composed of N color histogramsThe normalized feature histogram h extracted by the selected tracking target image in the initialization stage0To replace the N histograms in the original model.
Optionally, according to the foregoing embodiment of the present application, in step S108, updating the feature model according to the tracking result of the current frame image includes:
step S1081, performing feature extraction on the image region of the obtained tracking result of the current frame image, and performing normalization processing to obtain a plurality of corresponding feature information.
And step S1083, acquiring a preset probability.
Step S1085, replacing any one of the feature information in the feature model with a preset probability through preset feature information in the feature information corresponding to the obtained current frame image, so as to update the feature model.
According to the scheme, the feature model is updated by replacing any information in the feature model with the preset probability through the feature information of the current frame image, so that the feature model can change along with the change of the tracked target, on one hand, the latest target feature can be ensured to be introduced into the new model, and on the other hand, the introduced random performance can effectively keep the features of the target at all times in the historical process of tracking, so that the diversity of the feature information in the model is ensured, the robustness of the model is improved, and the influence of the environment, light rays and the like in the video on tracking is reduced as much as possible.
Optionally, according to the above embodiment of the present application, in step S1083, the obtaining of the preset probability includes:
step S1083a, acquiring the babbitt distance between the feature information of the current frame image and the feature model updated most recently.
Step S1083b, obtaining a preset probability by the following formula:
wherein p is a predetermined probability, dmedianBabbitt for feature information and most recently updated feature modelThe distance, σ, is a predetermined constant.
Specifically, the preset constant σ is used to determine the probability of control update.
Optionally, according to the above embodiment of the present application, in step S1083a, obtaining the babbitt distance between the feature information of the current frame image and the feature model that is updated last time includes: and determining the median value of the plurality of Barcol distances between the feature information of the current frame image and the plurality of feature information in the latest updated model as the Barcol distance between the feature information and the latest updated model.
In an alternative embodiment, the feature information of the current frame image is taken as htFor example, h can be computed one by onetPasteur distance from N histograms yields di1,2,3 …, N, and diArranging in ascending order, and taking the median dmedianAs htDistance from the model.
Further, in the process of determining whether the tracking target is lost, the 1-d obtained by the above calculation may be usedmedianAs color confidence; in the step of retrieving the tracked target, the method can be used for calculating the similarity between the candidate target and the tracked target to select the candidate target; in the scheme, the change of the target appearance caused by the change of environment, illumination and the like in the long-time target tracking process is considered, and N feature vectors are adopted to express the features of the target in different environments; during updating, a random replacement mode with a certain probability is adopted, so that the difference of N characteristic vectors in the model is ensured, the historical information of the target is kept, and the robustness of the long-time tracking system can be effectively improved. The model is not limited to specific color features, and may use the simplest color histogram or a complex feature vector calculation method.
Optionally, according to the above embodiment of the present application, in step S104, extracting feature information from an image region of a tracking target includes:
step S1041, removing the background image in the image region of the tracking target.
In step S1043, the image area without the background image is divided into a plurality of images along a preset direction.
Specifically, the preset direction may be determined according to a preset tracking target, and taking the tracking target as a walking person as an example, since the walking person is usually in an upright shape, the preset direction may be a vertical direction.
Step S1045, obtaining feature information of the plurality of equally divided images.
Step S1047, connecting the feature information of the plurality of images after being evenly divided according to the order of division, to obtain the feature information of the image of the tracking target.
In an alternative embodiment, as shown in the above embodiment, the image feature used in the modeling process is a Color feature, specifically, a Color Name histogram, and before the Color Name histogram is calculated, salience Segmentation is performed on the image to remove the interference of the background; taking a tracking object as a walking person as an example, and aiming at the characteristic that most pedestrians are in an upright state, dividing an image obtained after main component segmentation into M equal parts in the vertical direction before calculating a histogram, and independently counting the histogram for each image; and after connecting the M image histograms in sequence, normalizing the M image histograms to serve as color feature information.
Optionally, according to the foregoing embodiment of the present application, the feature information is image color feature information, where the image color feature information includes: color name information and/or hue information.
Example 2
According to an embodiment of the present invention, there is provided an embodiment of an object re-recognition apparatus, and fig. 2 is a schematic diagram of an object re-recognition apparatus according to an embodiment of the present invention, as shown in fig. 2, the apparatus including:
the acquisition module 10 is configured to acquire a tracking target and an image area of the tracking target.
Specifically, the tracking target may be a target specified by an operator or determined by a pedestrian detector, and the image area of the tracking target may be an area containing the tracking target indicated in an image of a certain frame of the video by an operator or an image area determined by a pedestrian detector in a certain frame of the video.
And a constructing module 20, configured to extract feature information from the image region of the tracking target, and construct a feature model according to the feature information.
Specifically, the extracted features may be color features, edge features, and the like of an image, and since a tracking target is usually dynamic in a video, tracking only using the shape of the tracking target as a model has certain difficulty and is low in accuracy, but generally for continuous images in the video, the tracking target changes in time and the shape changes with the change of a timestamp, but the features of the images are usually consistent, and therefore the above steps construct the model by using the extracted image features.
The determining module 30 is configured to determine a tracking state of the tracking target according to a reliability of a tracking result of the current frame image, where the reliability of the tracking result is determined by a similarity between feature information of a preset region of the current frame image and the feature model.
Specifically, the tracking result includes a region and a reliability of the tracking target in the image, and the tracking state of the tracking target may include three states of non-loss, low reliability, and loss. In an alternative embodiment, a confidence threshold may be set, and if the confidence level of the tracking result is determined to exceed the preset confidence threshold, the tracking result is determined not to be lost. When the reliability of the tracking result is determined according to the similarity between the feature information of the preset region of the current image and the feature model, the reliability can be determined by using the first feature information of the image, or by using a plurality of feature information of the image after information fusion.
And the updating module 40 is used for updating the feature model according to the tracking result of the current frame image under the condition that the tracking target is determined not to be lost according to the tracking state.
It should be noted here that, because the specific task may be in a dynamic state in the video, and other environmental information in the video also changes over time, that is, the shape of the specific task changes constantly in the video, and the illumination and environment in the video also change, it is very difficult to perform tracking or re-finding simply by tracking the shape of the target, and further, an accurate result cannot be obtained by performing tracking by always using the initially determined feature model of the tracked target, and therefore, the feature model of the tracked target introduced in the above scheme can effectively remove the influence of the change in the environment or the change in the shape of the tracked target in the tracking or re-finding process, thereby upgrading the robustness of the tracking model.
According to the scheme, the tracking target and the image area of the tracking target are obtained through the obtaining module, the feature information is extracted from the image area of the tracking target through the constructing module, the feature model is constructed according to the feature information, the tracking state of the tracking target is determined according to the reliability of the tracking result of the current frame image through the determining module, and the feature model is updated according to the tracking result of the current frame image through the updating module under the condition that the tracking target is determined not to be lost according to the tracking state. According to the scheme, the characteristic model is constructed according to the characteristic information of the image area of the opera tracking target, the characteristic model is continuously updated according to different tracking results in the tracking process, and the characteristic model is used as the tracking model for tracking, so that the robustness of the tracking model is improved, and the technical problem of poor robustness of the re-identification technology of the tracking target in the existing tracking technology is solved.
Optionally, according to the above embodiment of the present application, the above configuration module includes:
and the initialization submodule is used for replacing original characteristic information in a preset model by the characteristic information extracted from the image area of the tracking target to obtain a characteristic model.
Optionally, according to the above embodiment of the present application, the update module includes:
the extraction submodule is used for extracting the characteristics of an image area for acquiring the tracking result of the current frame image and carrying out normalization processing to obtain a plurality of corresponding characteristic information;
the first obtaining submodule is used for obtaining a preset probability;
and the replacing submodule is used for replacing any one piece of characteristic information in the characteristic model with a preset probability through the preset characteristic information in the acquired characteristic information corresponding to the current frame image so as to update the characteristic model.
Optionally, according to the above embodiment of the present application, the first obtaining sub-module includes:
the acquisition unit is used for acquiring the Pasteur distances between a plurality of characteristic information of the current frame image and the latest updated characteristic model;
the calculating unit is used for acquiring the preset probability through the following formula:
wherein p is a predetermined probability, dmedianσ is a preset constant, which is the babbitt distance between the feature information and the feature model updated most recently.
Optionally, according to the above embodiment of the present application, the obtaining unit includes:
and the determining subunit is used for determining that the median of the plurality of barbituric distances between the feature information of the current frame image and the most recently updated model is the barbituric distance between the plurality of feature information and the most recently updated model.
Optionally, according to the above embodiment of the present application, the above configuration module includes:
the background removing submodule is used for removing a background image in an image area of the tracking target;
the segmentation submodule is used for segmenting the image area without the background image into a plurality of images along a preset direction;
the second obtaining submodule is used for obtaining the characteristic information of the plurality of equally divided images;
and the connecting sub-module is used for connecting the feature information of the plurality of images after the average division according to the dividing sequence to obtain the feature information of the image of the tracking target.
Optionally, according to the foregoing embodiment of the present application, the feature information is image color feature information, where the image color feature information includes: color name information and/or hue information.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units may be a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.
Claims (13)
1. A method of object re-identification, comprising:
acquiring a tracking target and an image area of the tracking target;
extracting characteristic information from the image area of the tracking target, and constructing a characteristic model according to the characteristic information;
determining the tracking state of a tracking target according to the reliability of the tracking result of the current frame image, wherein the reliability of the tracking result is determined by the similarity between the feature information of the preset area of the current frame image and the feature model;
under the condition that the tracking target is determined not to be lost according to the tracking state, updating the feature model according to the tracking result of the current frame image;
wherein, updating the feature model according to the tracking result of the current frame image comprises:
performing feature extraction on an image area of the obtained tracking result of the current frame image, and performing normalization processing to obtain corresponding feature information;
acquiring a preset probability;
and replacing any one characteristic information in the characteristic model with a preset probability through preset characteristic information in the acquired characteristic information corresponding to the current frame image so as to update the characteristic model.
2. The method of claim 1, wherein constructing a feature model from the feature information comprises:
and replacing original characteristic information in a preset model by the characteristic information extracted from the image area of the tracking target to obtain the characteristic model.
3. The method of claim 1, wherein obtaining the predetermined probability comprises:
acquiring the Pasteur distance between the feature information of the current frame image and the feature model which is updated last time;
obtaining the preset probability through the following formula:
wherein p is the preset probability, dmedianAnd the sigma is a preset constant and is the Pasteur distance between the feature information and the feature model which is updated last time.
4. The method of claim 3, wherein obtaining the Bhattacharyya distance between the feature information of the current frame image and the most recently updated feature model comprises:
determining a median value of a plurality of Barcol distances between the feature information of the current frame image and the feature information in the latest updated model as the Barcol distance between the feature information and the latest updated model.
5. The method according to any one of claims 1 to 4, wherein extracting feature information from the image region of the tracking target comprises:
removing a background image in an image area of the tracking target;
dividing the image area without the background image into a plurality of images along a preset direction;
acquiring the feature information of the plurality of equally divided images;
and connecting the equally divided characteristic information of the plurality of images according to a segmentation order to obtain the characteristic information of the image of the tracking target.
6. The method according to any one of claims 1 to 4, wherein the feature information is image color feature information, wherein the image color feature information includes: color name information and/or hue information.
7. An object re-recognition apparatus, comprising:
the system comprises an acquisition module, a tracking module and a processing module, wherein the acquisition module is used for acquiring a tracking target and an image area of the tracking target;
the construction module is used for extracting characteristic information from the image area of the tracking target and constructing a characteristic model according to the characteristic information;
the determining module is used for determining the tracking state of the tracking target according to the reliability of the tracking result of the current frame image, wherein the reliability of the tracking result is determined by the similarity between the feature information of the preset area of the current frame image and the feature model;
the updating module is used for updating the feature model according to the tracking result of the current frame image under the condition that the tracking target is determined not to be lost according to the tracking state;
the update module includes:
the extraction submodule is used for extracting the characteristics of an image area for acquiring the tracking result of the current frame image and carrying out normalization processing to obtain corresponding characteristic information;
the first obtaining submodule is used for obtaining a preset probability;
and the replacing submodule is used for replacing any one piece of characteristic information in the characteristic model with a preset probability through the preset characteristic information in the acquired characteristic information corresponding to the current frame image so as to update the characteristic model.
8. The apparatus of claim 7, wherein the configuration module comprises:
and the initialization submodule is used for replacing original characteristic information in a preset model by the characteristic information extracted from the image area of the tracking target to obtain the characteristic model.
9. The apparatus of claim 7, wherein the first acquisition submodule comprises:
the acquisition unit is used for acquiring the Bhattacharyya distance between the feature information of the current frame image and the feature model which is updated at the latest time;
a calculating unit, configured to obtain the preset probability according to the following formula:
wherein p is the preset probability, dmedianAnd the sigma is a preset constant and is the Pasteur distance between the feature information and the feature model which is updated last time.
10. The apparatus of claim 9, wherein the obtaining unit comprises:
and the determining subunit is used for determining that the median value of the plurality of barbituric distances between the feature information of the current frame image and the plurality of feature information in the latest updated model is the barbituric distance between the feature information and the latest updated model.
11. The apparatus of any one of claims 7 to 10, wherein the construction module comprises:
the background removing submodule is used for removing a background image in the image area of the tracking target;
the segmentation submodule is used for segmenting the image area with the background image removed into a plurality of images along a preset direction;
the second obtaining submodule is used for obtaining the feature information of the evenly divided images;
and the connecting sub-module is used for connecting the equally divided characteristic information of the plurality of images according to the dividing sequence to obtain the characteristic information of the image of the tracking target.
12. The apparatus according to any one of claims 7 to 10, wherein the feature information is image color feature information, wherein the image color feature information includes: color name information and/or hue information.
13. The apparatus according to claim 11, wherein the feature information is image color feature information, wherein the image color feature information includes: color name information and/or hue information.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611265859.0A CN108268823B (en) | 2016-12-30 | 2016-12-30 | Target re-identification method and device |
| PCT/CN2017/116330 WO2018121287A1 (en) | 2016-12-30 | 2017-12-15 | Target re-identification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611265859.0A CN108268823B (en) | 2016-12-30 | 2016-12-30 | Target re-identification method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108268823A CN108268823A (en) | 2018-07-10 |
| CN108268823B true CN108268823B (en) | 2021-07-20 |
Family
ID=62710168
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611265859.0A Active CN108268823B (en) | 2016-12-30 | 2016-12-30 | Target re-identification method and device |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN108268823B (en) |
| WO (1) | WO2018121287A1 (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034109B (en) * | 2018-08-16 | 2021-03-23 | 新智数字科技有限公司 | Pedestrian re-identification method and device based on clustering algorithm |
| CN111127508B (en) * | 2018-10-31 | 2023-05-02 | 杭州海康威视数字技术股份有限公司 | Target tracking method and device based on video |
| CN109635657B (en) * | 2018-11-12 | 2023-01-06 | 平安科技（深圳）有限公司 | Target tracking method, device, equipment and storage medium |
| CN111753601B (en) * | 2019-03-29 | 2024-04-12 | 华为技术有限公司 | Image processing method, device and storage medium |
| CN110907909B (en) * | 2019-10-30 | 2023-09-12 | 南京市德赛西威汽车电子有限公司 | Radar target identification method based on probability statistics |
| CN112784648B (en) * | 2019-11-07 | 2022-09-06 | 中国科学技术大学 | Method and device for optimizing feature extraction of pedestrian re-identification system of video |
| CN111709978A (en) * | 2020-05-06 | 2020-09-25 | 广东康云科技有限公司 | Cross-screen target tracking method, system, device and storage medium |
| CN111738063B (en) * | 2020-05-08 | 2023-04-18 | 华南理工大学 | Ship target tracking method, system, computer equipment and storage medium |
| CN111860190B (en) * | 2020-06-24 | 2024-04-12 | 国汽(北京)智能网联汽车研究院有限公司 | Method, device, equipment and storage medium for target tracking |
| CN113536914A (en) * | 2021-06-09 | 2021-10-22 | 重庆中科云从科技有限公司 | Object tracking identification method, system, equipment and medium |
| CN113743262A (en) * | 2021-08-23 | 2021-12-03 | 浙江大华技术股份有限公司 | Target area vehicle tracking method and device, computer equipment and storage medium |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6590999B1 (en) * | 2000-02-14 | 2003-07-08 | Siemens Corporate Research, Inc. | Real-time tracking of non-rigid objects using mean shift |
| CN101888479A (en) * | 2009-05-14 | 2010-11-17 | 汉王科技股份有限公司 | Method and device for detecting and tracking target image |
| CN102339390A (en) * | 2011-09-16 | 2012-02-01 | 天津市亚安科技股份有限公司 | Method and system for updating target template of video monitoring system |
| CN104020466A (en) * | 2014-06-17 | 2014-09-03 | 西安电子科技大学 | Maneuvering target tracking method based on variable structure multiple models |
| CN104036523A (en) * | 2014-06-18 | 2014-09-10 | 哈尔滨工程大学 | Improved mean shift target tracking method based on surf features |
| CN104240266A (en) * | 2014-09-04 | 2014-12-24 | 成都理想境界科技有限公司 | Target object tracking method based on color-structure features |
| CN104252178A (en) * | 2014-09-12 | 2014-12-31 | 西安电子科技大学 | Strong maneuver-based target tracking method |
| CN105335986A (en) * | 2015-09-10 | 2016-02-17 | 西安电子科技大学 | Characteristic matching and MeanShift algorithm-based target tracking method |
| CN105931269A (en) * | 2016-04-22 | 2016-09-07 | 海信集团有限公司 | Tracking method for target in video and tracking device thereof |
| CN106023155A (en) * | 2016-05-10 | 2016-10-12 | 电子科技大学 | Online object contour tracking method based on horizontal set |
| CN106056627A (en) * | 2016-05-30 | 2016-10-26 | 河海大学 | Robustness object tracking method based on local identification sparse representation |
| CN106097385A (en) * | 2016-05-31 | 2016-11-09 | 海信集团有限公司 | A kind of method and apparatus of target following |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101901486B (en) * | 2009-11-17 | 2012-10-03 | 华为技术有限公司 | Method for detecting moving target and device thereof |
| CN105654508B (en) * | 2015-12-24 | 2018-06-01 | 武汉大学 | Monitor video method for tracking moving target and system based on adaptive background segmentation |
| CN105930833B (en) * | 2016-05-19 | 2019-01-22 | 重庆邮电大学 | A kind of vehicle tracking and dividing method based on video monitoring |
-
2016
- 2016-12-30 CN CN201611265859.0A patent/CN108268823B/en active Active
-
2017
- 2017-12-15 WO PCT/CN2017/116330 patent/WO2018121287A1/en active Application Filing
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6590999B1 (en) * | 2000-02-14 | 2003-07-08 | Siemens Corporate Research, Inc. | Real-time tracking of non-rigid objects using mean shift |
| CN101888479A (en) * | 2009-05-14 | 2010-11-17 | 汉王科技股份有限公司 | Method and device for detecting and tracking target image |
| CN102339390A (en) * | 2011-09-16 | 2012-02-01 | 天津市亚安科技股份有限公司 | Method and system for updating target template of video monitoring system |
| CN104020466A (en) * | 2014-06-17 | 2014-09-03 | 西安电子科技大学 | Maneuvering target tracking method based on variable structure multiple models |
| CN104036523A (en) * | 2014-06-18 | 2014-09-10 | 哈尔滨工程大学 | Improved mean shift target tracking method based on surf features |
| CN105405154A (en) * | 2014-09-04 | 2016-03-16 | 成都理想境界科技有限公司 | Target object tracking method based on color-structure characteristics |
| CN104240266A (en) * | 2014-09-04 | 2014-12-24 | 成都理想境界科技有限公司 | Target object tracking method based on color-structure features |
| CN104252178A (en) * | 2014-09-12 | 2014-12-31 | 西安电子科技大学 | Strong maneuver-based target tracking method |
| CN105335986A (en) * | 2015-09-10 | 2016-02-17 | 西安电子科技大学 | Characteristic matching and MeanShift algorithm-based target tracking method |
| CN105931269A (en) * | 2016-04-22 | 2016-09-07 | 海信集团有限公司 | Tracking method for target in video and tracking device thereof |
| CN106023155A (en) * | 2016-05-10 | 2016-10-12 | 电子科技大学 | Online object contour tracking method based on horizontal set |
| CN106056627A (en) * | 2016-05-30 | 2016-10-26 | 河海大学 | Robustness object tracking method based on local identification sparse representation |
| CN106097385A (en) * | 2016-05-31 | 2016-11-09 | 海信集团有限公司 | A kind of method and apparatus of target following |
Non-Patent Citations (2)
| Title |
|---|
| "基于前景概率函数的目标跟踪";周斌等;《北京理工大学学报》;20110430;第31卷(第4期);第443页第3.2节 * |
| "自适应分块的多特征融合多目标跟踪";别秀德等;《西安电子科技大学学报（自然科学版）》;20160929;第44卷(第2期);第152-154页第2节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2018121287A1 (en) | 2018-07-05 |
| CN108268823A (en) | 2018-07-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108268823B (en) | Target re-identification method and device | |
| US11302315B2 (en) | Digital video fingerprinting using motion segmentation | |
| CN108229456B (en) | Target tracking method and device, electronic equipment and computer storage medium | |
| JP6619634B2 (en) | Image processing apparatus and image processing method | |
| Noh et al. | A new framework for background subtraction using multiple cues | |
| US10121068B2 (en) | Information processing apparatus, information processing method and program | |
| CN107346409B (en) | pedestrian re-identification method and device | |
| Yang et al. | Robust superpixel tracking | |
| EP3373248A1 (en) | Method, control device, and system for tracking and photographing target | |
| JP6062791B2 (en) | License plate character segmentation using likelihood maximization | |
| US20160092727A1 (en) | Tracking humans in video images | |
| US20160070976A1 (en) | Image processing apparatus, image processing method, and recording medium | |
| US20070154066A1 (en) | Object tracking systems and methods | |
| US10007678B2 (en) | Image processing apparatus, image processing method, and recording medium | |
| US9105109B2 (en) | Method for superpixel life cycle management | |
| JP6947508B2 (en) | Moving object detection device, moving object detection system, and moving object detection method | |
| WO2019197021A1 (en) | Device and method for instance-level segmentation of an image | |
| WO2017168462A1 (en) | An image processing device, an image processing method, and computer-readable recording medium | |
| Khorrami et al. | Multiple animal species detection using robust principal component analysis and large displacement optical flow | |
| JP2017033372A (en) | Person recognition device and program therefor | |
| CN113963303A (en) | Image processing method, video recognition method, device, equipment and storage medium | |
| JP2017084006A (en) | Image processor and method thereof | |
| Meenatchi et al. | Multiple object tracking and segmentation in video sequences | |
| Xu et al. | Unsupervised saliency estimation based on robust hypotheses | |
| KR101489216B1 (en) | Apparatus and method for recognizing a place based on a straight line |
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 |