TWI517100B - Method for tracking moving object and electronic apparatus using the same - Google Patents

Description

Moving object tracking method and electronic device

The present invention relates to an object recognition method and an electronic device, and more particularly to a moving object tracking method and an electronic device.

With the widespread use of Surveillance Equipment, a variety of intelligent surveillance systems have emerged. Single cameras can be used to analyze simple human behavioral activities, but only for small-scale surveillance applications. Combined with multiple cameras, it can be used to analyze a wide range of behavioral activities, such as pedestrian walking path analysis, store customer shopping behavior analysis and abnormal behavior analysis.

The erection of multiple cameras can be divided into an Overlapping Field and a Non-overlapping Field. Generally speaking, the target tracking of multi-camera is performed in the overlapping area, and the absolute position in the space is the main discriminative feature. By using the corrected camera parameters and the three-dimensional scene model, the absolute value of the target in the space can be estimated. Position to determine whether the target is the same tracking object. However, due to the calculation of the amount of calculation and the cost of the economy, it is unlikely that the monitoring images of all the cameras overlap in practical applications; otherwise, the field of view has no overlapping environment. The multi-camera configuration is more flexible, lower cost, and has a wider monitoring range. However, due to the spatial discontinuity and the difference between the camera placement angle and the environment, the camera captures the object due to external conditions. Influence, which in turn causes difficulties in object matching.

In view of this, the present invention provides a moving object tracking method and an electronic device capable of performing long-range tracking of a wide range of moving objects.

The invention provides a moving object tracking method comprising the following steps. Receiving video data from a plurality of cameras, wherein the camera includes a first camera having a first range of viewing angles and a second camera having a second range of viewing angles, the video data including a first shooting angle range and a first The first video data and the second video data in the range of the viewing angle are not overlapped, and the first shooting angle range does not overlap with the second shooting angle range. Detecting at least one first foreground object and at least one second foreground object from the first video data and the second video data, and tracking the first foreground object and the second foreground object to obtain each of the first a moving path of the foreground object and a moving path of each of the second foreground objects. Estimating the movement time of the first foreground object from the first shooting angle range to the second shooting angle range, and calculating the brightness relationship of the first video material and the second video material. And comparing the first foreground object and the second foreground object according to the moving time, the color information of the first foreground object, and the color information of the second foreground object, The first associated with the second foreground object A foreground object, thereby establishing a complete moving path of each of said second foreground objects.

In an embodiment of the invention, the step of detecting the first foreground object and the second foreground object from the first video data and the second video data respectively comprises: using a Gaussian mixture model to respectively establish corresponding to the first a first background model and a second background model of the second shooting angle range; and the background subtraction method, the first background image and the second background model are respectively obtained from the first video data and the second video data respectively a first shadow foreground object and at least a second shadow foreground object; and performing shadow filtering processing and type processing on the first shadow foreground object and the second shadow foreground object to respectively generate the first a foreground object and the second foreground object.

In an embodiment of the invention, the step of tracking the first foreground object and the second foreground object to obtain a moving path of each of the first foreground objects and a moving path of each of the second foreground objects The method includes: determining whether an overlap occurs between the first reference foreground object of the first foreground object and the first foreground object in the first picture of the first video material; when it is determined that an overlap occurs Calculating a first predicted position of the first reference foreground object in a next picture of the first picture according to the Kalman filter, and performing a mean shift algorithm for the first predicted position to obtain the first detected position, and then according to The Kalman filter corrects the first detection position to obtain the first correction position; when it is determined that no overlap occurs, the first foreground object is tracked according to the block tracking method; and the second video data is determined to be the second Whether a overlap occurs between the second reference foreground object of the second foreground object and one of the second foreground objects in the picture; When an overlap occurs, a second reference foreground object is calculated according to a Kalman filter to a second predicted position in a next picture of the second picture, and a mean shift algorithm is performed for the second predicted position to obtain a second detected position And correcting the second detection position according to the Kalman filter to obtain the second correction position; and when determining that no overlap occurs, tracking the second foreground object according to the block tracking method.

In an embodiment of the invention, the estimating the first foreground object from The moving the first shooting angle range to the moving time between the second shooting angle range, and calculating the brightness relationship of the first video data and the second video data comprises: counting a plurality of known moving objects from the training phase a time required for the first shooting angle range to move between the second shooting angle range to obtain the moving time; and during the training phase, using the brightness conversion function, according to the known moving object respectively at the first shooting angle The range and the color information of the second shooting angle range are used to obtain the brightness relationship.

In an embodiment of the invention, the second foreground object comprises a second mesh Marking the foreground object, and comparing the first foreground object and the second foreground object according to the moving time, the color information of the first foreground object, and the color information of the second foreground object The step of identifying the first foreground object associated with each of the second foreground objects to establish a complete moving path of each of the second foreground objects comprises: when detecting that the second target foreground object enters the second shooting angle range, according to Converting the color information of the second target foreground object according to the moving time; querying the database according to the moving time to obtain at least one first time-similar foreground object associated with the second target foreground object, wherein the database records each of the first objects Foreground object Corresponding time point and color information, the first time-similar foreground object is a first foreground object in the time point that meets the moving time; the color information according to the second target foreground object and the first time is similar Color information of the foreground object, comparing the second target foreground object with each of the first time foreground objects, to identify the first target foreground object associated with the second target foreground object from each of the first time-similar foreground objects The color information of the first target foreground object is similar to the second target foreground object; and the complete moving path of the second target foreground object is established according to the moving path of the first target foreground object and the moving path of the second target foreground object.

The invention further provides an electronic device comprising a storage unit and one or more processing units, wherein the processing unit is coupled to the storage unit. The storage unit is used to record a plurality of modules. The processing unit is configured to access and execute the module recorded in the storage unit. The module includes: a data receiving module, a detecting module, a tracking module, and an identification module. The data receiving module is configured to receive video data from a plurality of cameras respectively, wherein the camera comprises a first camera having a first shooting angle range and a second camera having a second shooting angle range, the video data respectively corresponding to the first camera The first video data and the second video data of the range of the viewing angle and the second shooting angle range are captured, and the first shooting angle range does not overlap with the second shooting angle range. The detecting module is configured to detect at least one first foreground object and at least one second foreground object from the first video data and the second video data respectively. The tracking module is configured to track the first foreground object and the second foreground object to obtain a moving path of each of the first foreground objects and a moving path of each of the second foreground objects. The identification module is configured to estimate that the first foreground object moves from the first shooting angle range to the second shooting angle Calculating a brightness relationship between the first video data and the second video data, and calculating, according to the moving time, color information of the first foreground object, and color information of the second foreground object, And identifying, by the first foreground object and the second foreground object, a first foreground object associated with each of the second foreground objects, thereby establishing a complete moving path of each of the second foreground objects.

In an embodiment of the invention, the detecting module utilizes Gaussian mixing a first background model and a second background model corresponding to the first shooting angle range and the second shooting angle range, respectively, and the background subtraction method, respectively, according to the first background model and the second background model, respectively, from the first video And the second video material acquires at least one first shadow foreground object and at least one second shadow foreground object, and performs shadow filtering processing and pattern processing on the first shadow foreground object and the second shadow foreground object And generating the first foreground object and the second foreground object, respectively.

In an embodiment of the invention, the tracking module determines the first Whether the first reference foreground object of the first foreground object overlaps with one of the first foreground objects in the first picture of the video material; when it is judged that the overlap phenomenon occurs, the calculation is performed according to the Kalman filter The first reference foreground object is at a first predicted position in a next picture of the first picture, and performs a mean shift algorithm for the first predicted position to obtain a first detected position, and then corrects the first detect according to the Kalman filter The position is measured to obtain the first corrected position; when it is determined that no overlap occurs, the first foreground object is tracked according to the block tracking method. The tracking module further determines a second reference of the second foreground object in the second picture of the second video material Whether the overlap phenomenon occurs between one of the foreground object and the second foreground object; when it is determined that the overlap phenomenon occurs, calculating the second reference foreground object in the next picture of the second picture according to the Kalman filter Performing a mean shift algorithm for the second predicted position to obtain a second detected position, and then correcting the second detected position according to the Kalman filter to obtain a second corrected position; and when it is determined that no overlap occurs In the case of the phenomenon, the second foreground object is tracked according to the block tracking method.

In an embodiment of the present invention, the above-mentioned identification module counts the time required for a plurality of known moving objects to move from the first shooting angle range to the second shooting angle range during the training phase to obtain the movement. The time, and in the training phase, the brightness conversion function is used to obtain the brightness relationship according to the color information of the known moving object in the first shooting angle range and the second shooting angle range respectively.

In an embodiment of the invention, when the detecting module detects that the second target foreground object enters the second shooting angle range, the recognition module converts the color information of the second target foreground object according to the brightness relationship, and according to the moving Time, querying a database to obtain at least one first time-similar foreground object associated with the second target foreground object, wherein the first time-similar foreground object is the first foreground object in the time point that meets the moving time . The recognition module further compares the second target foreground object with each of the first time-similar foreground objects according to the color information of the second target foreground object and the color information of the foreground object at the first time, The first target foreground object associated with the second target foreground object is identified among the foreground objects, wherein the color information of the first target foreground object is similar to the second target foreground object. The recognition module is further based on the moving path of the first target foreground object and the second item A moving path of the foreground object, establishing a complete moving path of the second target foreground object.

Based on the above, the moving object tracking method and the electronic device proposed by the present invention first acquire the video data from the plurality of cameras, extract the foreground object corresponding to the moving object, and then track the foreground object. In addition, according to the time taken by the moving object to pass through the blind zone between the shooting field of view, and using the brightness conversion function to obtain the brightness relationship of the moving object in different shooting field ranges, to identify the same in different video materials. Move the object. Accordingly, the present invention can perform wide-area long-range tracking of moving objects under a plurality of cameras that do not overlap in the field of view, so as to achieve comprehensive security monitoring.

The above described features and advantages of the invention will be apparent from the following description.

100‧‧‧Electronic devices

110‧‧‧Processing unit

120‧‧‧ storage unit

122‧‧‧ Data receiving module

124‧‧‧Detection module

126‧‧‧Tracking module

128‧‧‧identification module

S202~S212, S506~S512‧‧‧ Flow of moving object tracking method

312~318, 326, 326a, 328, 410a~410b, 420a~420b‧‧‧ images

Blocks 422, 424, 502‧‧

504‧‧‧Listing

DB1, DB2‧‧‧ database

602‧‧‧ second target foreground object

S604~S618‧‧‧Mobile object recognition process

FIG. 1 is a block diagram of an electronic device according to an embodiment of the invention.

2 is a flow chart of a method for tracking a moving object according to an embodiment of the present invention.

3A and 3B are schematic diagrams showing a foreground object according to an embodiment of the present invention.

FIG. 4 is a schematic diagram of processing overlap phenomenon according to an embodiment of the present invention.

FIG. 5 is a schematic diagram of a moving object tracking process according to an embodiment of the invention.

FIG. 6 is a schematic diagram of a mobile object recognition process according to an embodiment of the present invention.

The components of the present invention will be described in detail in the following description in conjunction with the accompanying drawings. These examples are only a part of the invention and do not disclose all of the embodiments of the invention. Rather, these embodiments are merely examples of devices and methods within the scope of the patent application of the present invention.

1 is a block diagram of an electronic device according to an embodiment of the present invention, but is for convenience of description and is not intended to limit the present invention. FIG. 1 first introduces all the components and configuration relationships of the electronic device, and the detailed functions will be disclosed in conjunction with FIG. 2 .

Referring to FIG. 1 , the electronic device 100 of the embodiment is configured to perform long-range tracking on moving objects in the video data after receiving video data captured in a scene from different shooting angle ranges and different ambient light sources. The electronic device 100 can be implemented as a computer, a server, a distributed system, a smart phone, a tablet, any form of embedded system or device, and the invention is not limited to such implementations. The electronic device 100 includes a storage unit 110 and one or more processing units 120, the functions of which are described below.

The storage unit 110 is, for example, any type of fixed or removable random access memory, read only memory, flash memory, hard disk or the like or a combination of these devices for recording by the processing unit 120. A plurality of modules that can be loaded into the processing unit 120 to perform tracking of the moving object for the video material.

The processing unit 120 is, for example, a central processing unit, or other programmable general purpose or special purpose microprocessor, digital signal processor, and programmable. Controller, special application integrated circuit, programmable logic device or other similar device or a combination of these devices. The processing unit 120 is coupled to the storage unit 110, which can access and execute the modules recorded in the storage unit 110.

The module includes a data receiving module 112, a detecting module 114, and a tracking module. Group 116 and identification module 118. These modules are, for example, computer programs that can be loaded into the processing unit 120 to perform the functions of the moving object tracking method for the video material. The following is a detailed description of the detailed steps of the electronic device 100 performing the moving object tracking method.

2 is a moving object tracking method according to an embodiment of the invention Flow chart. Referring to FIG. 2, the method of the present embodiment is applied to the electronic device 100 of FIG. 1. Hereinafter, the detailed steps of the moving object tracking method of the present invention will be described with reference to various components in the electronic device 100.

First, the data receiving module 112 receives video resources from a plurality of cameras respectively. (Step S202). In detail, the data receiving module 112 can receive the video data captured by the data receiving module 112 from the camera by means of wired transmission or wireless transmission. The cameras herein have different ranges of viewing angles. Hereinafter, only the first camera and the second camera in the camera will be described. However, those skilled in the art can use the following steps to derive the steps of the moving object tracking method using the video data of other cameras.

In this embodiment, the first camera and the second camera respectively have the first A shooting angle range and a second shooting angle range, wherein the first shooting angle range and the second shooting angle range have no overlap. Here will be the first camera and the second The video data captured by the camera are defined as "first video data" and "second video data" respectively.

Then, the detecting module 114 is configured from the first video data and the second video data. At least one first foreground object and at least one second foreground object are respectively detected (step S204). In detail, in the first video data, the detection module 114 can use the Gaussian Mixture Model to establish a first background model for the scene monitored by the first camera, and obtain the video according to the first background model. The first foreground object in the data can then obtain the foreground image from the image by Background Subtraction. Taking FIG. 3A as an example, the image 312 is one of the first video data; the image 314 is the background image in the image 312, which is obtained according to the first background model; and the image 316 is the foreground mask of the image 312. (Foreground Mask); image 318 is the foreground image in image 312.

Image 326 in FIG. 3B is a partial enlarged view of image 316. Please refer to the figure 3B, the foreground object is obtained by the background subtraction method, and sometimes it is disturbed by the shadow and some noise (for example, the part 326a), and the outline of the foreground object cannot be completely extracted, and the foreground object including the shadow and the noise is included here. Defined as "first shadow foreground object". In this embodiment, the detecting module 114 can perform shadow removal and Morphological Operation on the first shadow foreground object to obtain a more complete foreground object, that is, the foregoing The first foreground object. The principle of shadow filtering is that when a pixel is covered by a shadow, its brightness (Brightness/Value) will decrease, but the hue will not change. Based on this feature, in this embodiment, a HSV (Hue, Saturation, Value) color model is used, The difference between the foreground image and the background image results in a difference in brightness, wherein the darkened portion can be regarded as a shadow, and the portion of the image is filled and eliminated by the morphology. For example, after the shadow 326 and the morphological processing of the image 326, the shadow portion of the portion 326a will be removed, resulting in an image 328 having a more complete foreground object. Similarly, in step S204 of FIG. 2, the detection module 114 also uses the Gaussian mixture model to establish a second background model for the scene monitored by the second camera, and obtains a second shadow in the video data according to the second background model. After the foreground object, the shadow filtering process and the morphological processing are used to generate the second foreground object, respectively. Please refer to the preceding paragraph for related instructions, which will not be repeated here.

Referring to FIG. 2 again, the detecting module 114 is detecting the first foreground object. And the second foreground object, the tracking module 116 tracks the first foreground object and the second foreground object to obtain a moving path of each of the first foreground objects and each of the second foreground objects The path of movement (Step S206). The tracking method used by the tracking module 116 is determined based on whether or not the foreground object overlaps in the video material. In detail, in this embodiment, when there is no overlap in the first foreground object, the tracking module 116 will use the block-based tracking algorithm to achieve the intersection of the first foreground object in the continuous image. The effect of tracking. On the other hand, when an overlap occurs in the first foreground object, the tracking module 116 can use a multi-object overlap tracking method, which is based on a Kalman filter combined with a Mean Shift Algorithm. Tracking the overlapping first foreground objects to solve the problem that the first foreground objects are mutually shielded. It should be noted that in other embodiments, the tracking module 116 may employ other tracking algorithms. The method tracks the moving object, and the present invention is not limited thereto. In order to facilitate the explanation of the mean shift algorithm and the Kalman filter applied to the multi-target overlap tracking method, only the "first reference foreground object" that is masked among the first foreground objects will be described below.

The tracking module 116 can store the color information of the first reference foreground object before the unmasking into a database (not shown) of the storage unit 110 by using a blob. In this embodiment, the color information is red information, green information, and blue information, and the color information may be stored in the database in the form of, for example, a RGB histogram. When the first reference foreground object produces a shadow on the first picture (ie, when the overlap phenomenon occurs), the tracking module 116 calculates the first reference foreground object in the next picture of the first picture using the Kalman filter. Predicting the position, and setting the first predicted position as the initial window of the mean shift algorithm, using the foreground mask to exclude the interference of the background pixel, obtaining the foreground image, performing back projection to obtain the image probability density map, and then adjusting the position of the search window. Until the convergence. Finally, the tracking module 116 then uses the Kalman filter to correct the first detection position to the first correction position.

Taking FIG. 4 as an example, the images 410a and 410b are shadows before the overlap phenomenon occurs. For example, the images 420a and 420b are images after the overlap phenomenon occurs. Here, block 422 may, for example, correspond to the first reference foreground object described above, and block 424 may, for example, correspond to another first foreground object. Since the tracking module 116 combines the Kalman filter and the mean shift algorithm, it can track the block 422 and the block 424 separately even if an overlap occurs. The tracking module 116 processes the first video data in the above manner to obtain a moving path of the first reference foreground object in the first video material. Similarly, the tracking module 116 also tracks the second in a similar manner. For foreground objects, please refer to the preceding paragraph for related explanations, which will not be repeated here.

In an embodiment, step S206 in FIG. 2 can further move the object of FIG. 5. The body tracking process is implemented. Referring to FIG. 5, when the detection module 114 detects a block blob 502 of a foreground object in the first video data or the second video data, the tracking module 116 stores the block blob 502 and the database. The foreground objects recorded in the tandem blobList 504 are paired (step S506), and it is judged whether or not the block blob 502 overlaps (step S508). When the tracking module 116 determines that an overlap occurs, the tracking module 116 tracks the block blob 502 using the mean shift algorithm and the Kalman filter (step S510). When the tracking module 116 determines that no overlap occurs, the tracking module 116 tracks the block blob 502 using a block-based tracking algorithm (step S512). The steps in Fig. 5 have been described in detail above, and will not be described again here.

Referring again to FIG. 2, steps S202 to S206 are mainly directed to different The video data is processed separately. Next, in a subsequent step, the recognition module 118 will align the moving objects that have taken the range of viewing angles across different cameras for complete tracking. The invention mainly utilizes the brightness relationship of the moving object under different shooting angle ranges, and predicts the time required for the moving object to traverse the blind zone between different scenes, thereby tracking the moving object.

In particular, the recognition module 118 will estimate the first foreground object from the first The shooting angle range is moved to a moving time between the second shooting angle ranges (step S208). In this embodiment, during the training phase, the recognition module 118 can count the time taken for a plurality of known moving objects to pass through the blind zone, and obtain a Gaussian distribution. (Gaussian Distribution) map. The identification module 118 can obtain the maximum possible time and the minimum possible time of the above moving time according to the Gaussian distribution map. For example, the maximum possible time may be the average of the above times plus one standard deviation; the minimum possible time may be the average of the above times minus one standard deviation.

Next, the recognition module 118 calculates a brightness relationship between the first video material and the second video data (step S210). In detail, due to the difference in the external ambient light source of each camera and the setting of the focal length and aperture of the camera itself, color differences in moving objects in different video materials may occur. Therefore, the recognition module 118 can use the brightness transfer function (Brightness Transfer Function) to perform color correction on the moving object, which is mainly corrected by converting the color information of one camera to the color information of another camera. The method of color.

In this embodiment, during the training phase, the recognition module 118 can use the brightness information to obtain the brightness relationship by using the color information of the known moving object in the first training video data and the second training video data respectively. . After obtaining the brightness relationship, the identification module 118 uses the color information in one of the first video data and the second video data as reference information, and uses the brightness relationship to convert the first video data and the second according to the reference information. Color information in the other of the video material. For example, if the identification module 118 uses the first video data as the reference information, the recognition module 118 converts the color information of the second video data by using the brightness conversion function, that is, the color information of the second foreground object will be Will be converted.

Then, the identification module 118 compares the first foreground object and the second foreground object according to the moving time, the color information of the first foreground object, and the color information of the second foreground object. And identifying a first foreground object associated with each of the second foreground objects, thereby establishing a complete moving path of each of the second foreground objects (step S212). In detail, when a moving object enters an image frame of the second video material and becomes a second foreground object (herein defined as a "second target foreground object"), the recognition module 118 can utilize the obtained brightness relationship pair. After color correction, the appearance model is established. In this embodiment, the identification module 118 uses the K-means Clustering Algorithm to divide the color information of the second target foreground object into, for example, 24 groups, and then by, for example, an equation ( 1) The normalized geometric distance is again classified for the color group: Wherein V ₁ and V ₂ are pixel values of different pixels, respectively, including red pixel values r ₁ and r ₂ , green pixel values g ₁ and g _{2 ,} and blue pixel values b ₁ and b ₂ . Thereafter, the recognition module 118 merges the groups of similar distances again until the cluster members no longer change. The recognition module 118 can establish an appearance model of the second target foreground object according to the last generated color group, for example, in the form of a histogram.

Then, the identification module 118 can select the second target according to the moving time. The foreground object is compared to the first foreground object. For example, when the second target foreground object enters the second video data, the recognition module 118 may calculate a time point of the second target foreground object in the first shooting angle range according to the moving time, and obtain a corresponding to the time point. A foreground object. In other words, the database in the storage unit 110 can record the time point and color information corresponding to each of the first foreground objects. knowledge The module 118 can query from the database, and obtain the first foreground object associated with the second target foreground object according to the estimated time point, which is defined as “the first time close foreground object”. Then, the recognition module 118 compares the similarity between the second target foreground object and each of the first time-similar foreground objects according to the second target foreground object and the color information of the first-time foreground object, and the similarity is high. The first foreground object that is adjacent to the foreground object is the first foreground object associated with the second target foreground object, and is defined herein as a "first target foreground object." Since the tracking module 116 has tracked the tracking path of the first target foreground object and the second target foreground object in step S206, the identification module 118 can obtain the second target foreground object in the first video data and the second video data. The full path of the move.

In an embodiment, steps S208 to S12 in FIG. 2 can be implemented by the moving object recognition process of FIG. 6. Referring to FIG. 6, when the detection module 114 detects the second target foreground object 602, the recognition module 118 determines whether the second target foreground object 602 is a new foreground object (step S604). In detail, in this embodiment, each foreground object includes a number in addition to the corresponding time point and color information. Therefore, in this step, the detection module 114 can query whether it is a new foreground object according to, for example, its number.

When the recognition module 118 determines that the second target foreground object 602 is not a new foreground object, the recognition module 118 stores the number, time point, and color information of the second target foreground object 602 in the storage unit 110 corresponding to the second camera. The second database DB2. On the contrary, when the recognition module 118 determines that the second target foreground object 602 is a new foreground object, the recognition module 118 will convert the second using the brightness conversion function. The color information of the target foreground object 602 (step S606), and according to the moving time, querying the first time-similar foreground object from the first database DB1 corresponding to the first camera in the storage unit 110 (step S607) to determine the first time Whether or not the foreground object is present (step S608).

When the identification module 118 queries the foreground object that is close to the first time, the identification The module 118 will calculate the similarity of the second target foreground object 602 to the foreground object that is similar to each of the first times (step S610). When the recognition module 118 cannot find the first foreground object with a similar time point, the recognition module 118 will also calculate the similarity between the second target foreground object 602 and the other first foreground objects (step S612).

Next, the recognition module 118 will query whether the first target foreground object associated with the second target foreground object 602 is correlated according to the similarity comparison result (step S614). When the recognition module 118 queries the first target foreground object, the number of the second target foreground object 602 is set to the number of the corresponding first foreground object, the matchedID 616, and the matchedID 620 is stored to the second database DB2. . When the recognition module 118 does not find the first target foreground object corresponding to the second target foreground object 602, the second target foreground object 602 will be given a new number newID 618.

In summary, the moving object tracking method and the electronic device proposed by the present invention first acquire the video object from the plurality of cameras, extract the foreground object corresponding to the moving object, and then track the foreground object. In addition, according to the time taken by the moving object to pass through the blind zone between the shooting field of view, and using the brightness conversion function to obtain the brightness relationship of the moving object in different shooting field ranges, to identify the same in different video materials. Move the object. Accordingly, the present invention can be used in multiple shots Under the camera with non-overlapping field of view, wide-area tracking of moving objects is performed to achieve comprehensive security monitoring.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and any one of ordinary skill in the art can make some changes and refinements without departing from the spirit and scope of the present invention. The scope of the invention is defined by the scope of the appended claims.

S202~S212‧‧‧ Flow of moving object tracking method

Claims (8)

A mobile object tracking method includes: receiving a plurality of video materials respectively from a plurality of cameras, wherein the camera includes a first camera having a first shooting angle range and a second camera having a second shooting angle range. The video data includes a first video data and a second video data respectively corresponding to the first shooting angle range and the second shooting angle range, and the first shooting angle range and the second shooting angle range do not overlap; The first video data and the second video data respectively detect at least one first foreground object and at least one second foreground object, and track the first foreground object and the second foreground object to obtain each of the first a moving path of the foreground object and a moving path of each of the second foreground objects, wherein the first foreground object and the second foreground object are tracked to obtain a moving path of each of the first foreground objects and each of the The step of moving the second foreground object includes: determining a first picture of the first video material, the first Whether a first reference foreground object of the foreground object and one of the first foreground objects overlap; when it is determined that the overlapping phenomenon occurs, calculating the first reference foreground object according to a Kalman filter a first predicted position in the next picture of the first picture, and performing a mean shift algorithm for the first predicted position to obtain a first detected position, and then correcting the first detect according to the Kalman filter Measuring position to obtain a first correction position; when judging that the overlap phenomenon does not occur, tracking according to a block tracking method Determining whether the first foreground object of the second foreground object overlaps with one of the second foreground object and the second foreground object in the second picture of the second video material a phenomenon; when it is determined that the overlapping phenomenon occurs, calculating a second predicted position of the second reference foreground object in a next picture of the second picture according to the Kalman filter, and performing the mean for the second predicted position a drift algorithm to obtain a second detection position, and then correcting the second detection position according to the Kalman filter to obtain a second correction position; and when determining that the overlap phenomenon does not occur, according to the block Tracking the second foreground object; estimating at least one moving time of the first foreground object moving from the first shooting angle range to the second shooting angle range, and calculating the first video material and the a brightness relationship of the second video material; and, according to the moving time, color information of the first foreground object, and color information of the second foreground object, The first object and the second foreground foreground object, according to the first foreground object to identify each of the second object associated with the foreground, to establish a complete path of movement of each of the second foreground object. The mobile object tracking method of claim 1, wherein the detecting the first foreground object and the second foreground object from the first video data and the second video data respectively comprises: utilizing a Gaussian mixture model, respectively corresponding to the first shooting angle range And a first background model of the second shooting angle range and a second background model; using a background subtraction method, according to the first background model and the second background model, respectively, from the first video data and the second Obtaining at least a first shadow foreground object and at least a second shadow foreground object; and performing a shadow filtering process and a pattern processing on the first shadow foreground object and the second shadow foreground object, The first foreground object and the second foreground object are generated separately. The moving object tracking method of claim 1, wherein the moving time of the first foreground object from the first shooting angle range to the second shooting angle range is estimated, and the first The step of the video data and the brightness relationship of the second video data includes: counting, during a training phase, a time required for moving a plurality of known moving objects from the first shooting angle range to the second shooting angle range Obtaining the moving time; and obtaining the brightness according to the color information of the first moving viewing angle range and the second shooting viewing angle range by the known moving object by using a brightness conversion function. relationship. The moving object tracking method according to claim 1, wherein the second foreground object includes a second target foreground object, and according to the moving time, color information of the first foreground object, and the Color information of the foreground object, which is compared with the first foreground object and the second foreground object The step of establishing the first foreground object associated with each of the second foreground objects to establish a complete moving path of each of the second foreground objects comprises: when detecting the second target foreground object entering the second shooting angle range Converting color information of the second target foreground object according to the brightness relationship; and querying a database according to the moving time to obtain at least one first time-similar foreground object associated with the second target foreground object, wherein the data The library records color information corresponding to each of the first foreground objects and a time point, wherein the first time-similar foreground object is a first foreground object in the time point that meets the moving time; according to the second target The color information of the foreground object and the color information of the foreground object in the first time are different from the foreground object in the first time and the foreground object in the first time. a first target foreground object associated with the second target foreground object, wherein the color information of the first target foreground object is similar The second target foreground object; and according to the path of the moving path of the first target object and the second foreground foreground target object, the second object to establish a complete movement path of the foreground object. An electronic device includes: a storage unit that records a plurality of modules; and one or more processing units coupled to the storage unit to access and execute the module recorded in the storage unit, the module The method includes: a data receiving module, respectively receiving a plurality of video materials from a plurality of cameras, wherein the camera includes a first camera having a first shooting angle range And a second camera of the second shooting angle range, the video data includes a first video material and a second video data respectively corresponding to the first shooting angle range and the second shooting angle range, Detecting at least one first foreground object and at least one second foreground object from the first video data and the second video data; Tracking module, tracking the first foreground object and the second foreground object to obtain a moving path of each of the first foreground objects and a moving path of each of the second foreground objects, wherein the tracking module determines In a first picture of the first video material, whether a first reference foreground object of the first foreground object and one of the first foreground objects overlap, when the tracking module determines that the occurrence occurs In the overlapping phenomenon, calculating a first predicted position of the first reference foreground object in a next picture of the first picture according to a Kalman filter, and the pin The first predicted position performs a mean shift algorithm to obtain a first detected position, and then corrects the first detected position according to the Kalman filter to obtain a first corrected position, when the tracking module determines When the overlapping phenomenon does not occur, the first foreground object is tracked according to a block tracking method, and the tracking module further determines a second foreground object in a second picture of the second video data. Whether the overlap occurs between the reference foreground object and one of the second foreground objects; When the tracking module determines that the overlapping phenomenon occurs, calculating a second predicted position of the second reference foreground object in the next picture of the second picture according to the Kalman filter, and performing for the second predicted position The mean shift algorithm is used to obtain a second detection position, and then the second detection position is corrected according to the Kalman filter to obtain a second correction position, and when the tracking module determines that the overlap does not occur Tracking the second foreground object according to the block tracking method; and identifying a module, estimating at least one of the first foreground object moving from the first shooting angle range to the second shooting angle range Moving time, calculating a brightness relationship of the first video material and the second video material, and comparing the color time according to the moving time, color information of the first foreground object, and color information of the second foreground object Determining the first foreground object and the second foreground object to identify the first foreground object associated with each of the second foreground objects, thereby establishing each of the The full path of movement of the foreground object. The electronic device of claim 5, wherein the detecting module uses a Gaussian mixture model to respectively establish a first background model corresponding to the first shooting angle range and the second shooting angle range, and a a second background model, the detection module uses a background subtraction method to obtain at least one first shadow foreground object from the first video data and the second video data according to the first background model and the second background model respectively And at least one second shadow foreground object, and the detection module is for the first shadow foreground object and the second shadow front The scene object performs a shadow filtering process and a type processing to generate the first foreground object and the second foreground object, respectively. The electronic device of claim 5, wherein the identification module is required to count a plurality of known moving objects from the first shooting angle range to the second shooting angle range during a training phase. Time to obtain the moving time, and the recognition module uses a brightness conversion function according to the brightness of the first moving viewing angle range and the second shooting angle range Information to obtain this brightness relationship. The electronic device of claim 5, further comprising a database for recording color information corresponding to each of the first foreground objects and a time point, the second foreground object comprising a second target foreground object, When the detecting module detects that the second target foreground object enters the second shooting angle range, the recognition module converts the color information of the second target foreground object according to the brightness relationship, and the Moving the time, querying the database to obtain at least one first time-similar foreground object associated with the second target foreground object, wherein the first time-similar foreground object is the first of the time points that meets the moving time a foreground object, the recognition module compares the color information of the foreground object of the second target with the color information of the foreground object of the first time, and compares the foreground object of the second target with the foreground object of the first time, Promising foreground objects from each of the first time Identifying a first target foreground object associated with the second target foreground object, wherein color information of the first target foreground object is similar to the second target foreground object, and the recognition module is based on the first target foreground object The moving path and the moving path of the second target foreground object establish a complete moving path of the second target foreground object.