Multi-target tracking method and face recognition method

[0036] As mentioned in the background technology section, the prior art ReID processing logic is fixed. After the target recognition processing is performed on the frame image, the position-related information is used to perform the first ReID. If the matching is not successful, the feature information is restarted. The second ReID. Because the time and place of image shooting are random, the light, angle, and posture are different, and the object is easily affected by factors such as detection accuracy and occlusion, the matching accuracy of this method is relatively low.

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0038] The method in this embodiment is mainly implemented by computer equipment; the computer equipment includes but is not limited to network equipment and user equipment. The network equipment includes, but is not limited to, a single network server, a server group composed of multiple network servers, or a cloud composed of a large number of computers or network servers based on Cloud Computing, where cloud computing is a type of distributed computing , A super virtual computer composed of a group of loosely coupled computer sets; the network where the computer device is located includes but is not limited to the Internet, wide area network, metropolitan area network, local area network, VPN network, etc. The user equipment includes, but is not limited to, PCs, tablet computers, smart phones, PDAs, IPTVs, etc.

[0039] It should be noted that the computer equipment and network are only examples. Other existing or future computer equipment or networks that can be applied to the present invention should also be included in the scope of protection of the present invention and included by reference. Here.

[0040] See figure 1 As shown, this embodiment provides a multi-target tracking method, which specifically includes the following steps:

[0041] Step S1, acquiring a video image to be processed, where the video image to be processed includes multiple frame images;

[0042] Step S2: Perform target recognition processing on each frame image to obtain position-related information and characteristic information of each target in each frame image, at least one of the frame images includes multiple targets;

[0043] Step S3: Analyze each target in each frame image to determine whether each target belongs to a specific target. When the target belongs to a specific target, perform step S4; when the target does not belong to a specific target, Go to step S5;

[0044] Step S4, perform re-identification processing based on the feature information or simultaneously based on the location-related information and the feature information, and perform step S8;

[0045] Step S5, first perform the first re-identification process according to the location-related information, and execute step S6;

[0046] Step S6, it is judged whether the matching is successful, when the matching fails, step S7 is executed; when the matching succeeds, step S8 is executed.

[0047] Step S7: Perform a second re-identification process according to the characteristic information;

[0048] Step S8, end.

[0049] After the target recognition process and before the re-recognition process, the present invention adds a step of analyzing and processing each target in each frame image to determine whether each target belongs to a specific target. Based on accuracy, considering that the specific target is not suitable for direct use Re-identification processing is performed on location-related information, so that when the target belongs to a specific target, re-identification processing is performed according to feature information or based on location-related information and feature information at the same time, so as to realize the dynamic ReID scheme for different objects and different situations in the video. The problem of low matching accuracy in the prior art improves the flexibility and accuracy of tracking matching.

[0050] First, step S1 is executed to obtain a video image to be processed.

[0051] The video image to be processed may be taken in real time, or any video image stored in advance, and its format is not limited.

[0052] The video image to be processed includes multiple frame images, and each frame image corresponds to a different time.

[0053] Then, step S2 is executed to perform target recognition processing on each frame image.

[0054] In this embodiment, methods such as R-CNN, Fast R-CNN, Faster R-CNN, R-FCN, YOLO, or SSD can be used to perform target recognition processing.

[0055] As a specific example, you can first identify the target category and position information (that is, the position of the target on the image, such as coordinates, size, etc.), and then obtain the speed, acceleration, etc. through judgment (for example, combined with multi-frame image tracking results) Other location-related information, and feature information is obtained through extraction processing during or after recognition.

[0056] It should be noted that the above examples are only to better illustrate the technical solutions of the present invention, rather than to limit the present invention. Those skilled in the art should understand that any target recognition that can obtain the position-related information and feature information of the target in the frame image All treatments should be included in the protection scope of the present invention.

[0057] The position-related information includes: position, size, speed, and other variables related to the position information of the target or its reciprocal. When the position-related information uses different variables, the target recognition processing can adopt different methods.

[0058] The feature information includes the feature vector of the object extracted by the neural network model, and it may also include the feature vector of the object extracted by other means, which is well known to those skilled in the art and will not be repeated here.

[0059] Next, step S3 is executed to analyze and process each target in each frame image to determine whether each target belongs to a specific target.

[0060] The inventor discovered through creative work: For a specific target, if the location-related information is directly used for the first re-identification process, even if the matching is unsuccessful, the feature information is used for the second re-identification process, which will eventually lead to accurate multi-target tracking. The rate is very low, and the existing two-step ReID processing can re-identify non-specific targets very accurately.

[0061] The analysis and processing in the present invention can use the position-related information between different targets in the same frame of image to determine whether the current frame belongs to a specific situation, or use the position-related information between the same target in different frame images to determine the target. Whether it belongs to a specific target, you can also combine the aforementioned two methods to determine whether the target belongs to a specific target. Both of them use position-related information between different targets to make judgments. Since the calculation of position-related information is relatively simple, the speed is relatively high. Fast, achieving high accuracy and high efficiency at the same time.

[0062] The following examples will be described in detail.

[0063] In the first example, the analysis processing may include: respectively calculating the overlap area ratio between two targets in the same frame of image, and when the overlap area ratio is greater than a first preset value, it is greater than the first preset value. The two targets corresponding to the overlap area ratio of the value are specific targets. When the overlapping area ratio is less than the first preset value, the two targets corresponding to the overlapping area ratios smaller than the first preset value are non-specific targets.

[0064] The inventor found through creative work that when the overlap area ratio between two objects in the image at the same moment is too large, the tracking and matching of the two objects will fail according to the position-related information X at this time, that is, the main target at this time is The matching failure is caused by the overlapping and interlacing of objects. For this reason, this embodiment adopts the method of horizontal comparison, that is, by calculating and judging the position-related information between two different targets in the image at the same time—the overlap area ratio, two objects with too large overlap area ratio are regarded as Specific targets are processed, and the rest are treated as non-specific targets.

[0065] Because the calculation and judgment of the overlap area ratio are simple and fast, both high accuracy and high efficiency are achieved.

[0066] It should be noted that when a target is determined as a specific target and a non-specific target in the above manner, the target is still treated as a specific target in the end. For example: when a target A and target B overlap area ratio is greater than the first preset value, but the overlap area ratio of target A and target C is less than the first preset value, then target A and target B are both specific targets, and Goal C is a non-specific goal.

[0067] It should be noted that for the case where the overlap area ratio is equal to the first preset value, the corresponding two targets can be judged as specific targets or non-specific targets, both of which are protected by the present invention. Within range.

[0068] The specific calculation of the overlap area ratio between the two targets may adopt various methods, which are well known to those skilled in the art, and will not be repeated here.

[0069] The value range of the first preset value in this embodiment may include: 0.3-0.8, such as 0.3, 0.4, 0.5, 0.55, 0.6, 0.7, or 0.8. The value of the first preset value cannot be too large, otherwise the constraints on the position-related information X are too small, and feature information cannot be introduced for re-identification under appropriate circumstances, and the advantages of the algorithm cannot be highlighted; the first preset The value of the value can not be too small, otherwise the constraint on the position-related information X is too large, resulting in two objects that are far apart also need to be identified by feature information calculation, which increases the calculation cost.

[0070] In the second example, the image in the i-1th frame includes M targets, and the image in the i-th frame includes N targets, and the analysis processing includes: calculating each target in the i-1th frame image and the i-th frame image The difference value of the position-related information of each target. When the difference between a target in the i-th frame image and the M targets in the i-1th frame image is greater than the second set value, the target is a specific target . Among them, i, M, and N are all positive integers, and the values ​​of M and N are different.

[0071] Through creative work, the inventor discovered that in certain application scenarios, such as when multiple people are tracked in a video and when the characters are interlaced or encountering, or when the target speed and path suddenly change in a sports game, according to the position Related information X matching is likely to go wrong. For this reason, this embodiment adopts the method of longitudinal comparison, that is, through the mutation of the position-related information X of the target in two adjacent images at different times, the target with too large abrupt change (that is, the current frame and the previous frame of image All targets whose position-related information has a difference greater than the second preset value) are treated as specific targets, and the rest are treated as non-specific targets.

[0072] Since the calculation of the difference of the position-related information X is simple and the speed is relatively fast, high accuracy and high efficiency are simultaneously achieved.

[0073] It should be noted that when the above method is adopted, the target D is judged only when the difference between the position-related information of the target D in the current frame and all (not part of) targets in the previous frame is greater than the second preset value For a specific goal.

[0074] It should be further explained that for the case where the difference is equal to the second preset value, either a processing method less than the second preset value or a processing method greater than the second preset value may be used, both of which are Within the protection scope of the present invention.

[0075] Since the position-related information X can include multiple parameters, the difference of the position-related information can also adopt a variety of ways, all of which are within the protection scope of the present invention, such as calculating the distance of the position, size, speed, etc., It can be the absolute distance of the position, the Euclidean distance of the position vector, or the cosine distance of the position vector.

[0076] When the distance is the cosine distance of the position vector (that is, the cosine distance of the vector change angle using position information as the vector), the value range of the second preset value may include 0.1-0.35, such as: 0.1, 0.15, 0.2, 0.25, 0.3 and 0.35 etc.

[0077] When the distance is the absolute distance of the position or the Euclidean distance of the position vector, the value range of the second preset value is directly related to the application scenario, for example, when the picture is a close-up, the distance between the targets can be compared Therefore, the value range of the second preset value increases accordingly; when the picture is a distant view, the distance between the targets can be relatively small, so that the value range of the second preset value decreases accordingly.

[0078] The value of the second preset value cannot be too large, otherwise the constraint on the position-related information X is too small and will not be effective, and feature information cannot be introduced at an appropriate time; the value of the second preset value is also It cannot be too small, otherwise the constraint on the location information X is too large, and the feature information calculation will be introduced with a high probability, increasing the calculation cost.

[0079] In the above two examples, only the position-related information between targets in the same frame or in different frames is used to determine a specific target. In addition, the method of this embodiment can also add a sensitive object recognition step to each frame of image, and then combine the position related information of the sensitive object to determine a specific target. For details, see the following three examples.

[0080] In the third example, the method further includes: recognizing sensitive objects in each frame of image, and obtaining location-related information of the sensitive objects; the analysis processing includes: calculating the relationship between each target in each frame of image and the For the spatial distance of the sensitive object in the frame image, the target corresponding to the spatial distance less than the third preset value is determined as a specific target.

[0081] The sensitive objects may include entrances and exits, obstacles and other objects.

[0082] The inventor discovered through creative work: In certain application scenarios, such as in the case of video tracking multi-person targets, if the camera tracking screen includes an entrance (such as: elevator entrance, building entrance, subway entrance, etc.), it often appears For a new person, the matching based on location-related information X is likely to go wrong. For this reason, this embodiment first identifies sensitive objects (such as entrances and exits), and then calculates the spatial distance between each target and the sensitive object. If the distance is too small, the target appears within a certain range of the entrance and exits, and you cannot rely on location-related information X Make a judgment, that is, judge a target that is too small in space with a sensitive object as a specific target.

[0083] Since sensitive objects can be identified while identifying the target, and the calculation of the spatial distance is simple and fast, high accuracy and high efficiency are achieved at the same time.

[0084] The spatial distance may be the absolute distance of the position, the Euclidean distance of the position vector, or the cosine distance of the position vector. It should be noted that, for the case where the spatial distance is equal to the third preset value, either a processing method less than the third preset value or a processing method greater than the third preset value may be used, all of which are Within the protection scope of the present invention.

[0085] Since the position-related information X can include multiple parameters, the difference of the position-related information can also adopt a variety of ways, all of which are within the protection scope of the present invention, such as calculating the distance of the position, size, speed, etc., It can be the absolute distance of the position, the Euclidean distance of the position vector, or the cosine distance of the position vector.

[0086] When the distance is the cosine distance of the position vector (that is, the cosine distance of the vector change angle using position information as the vector), the value range of the third preset value may include 0.1-0.35, such as: 0.1, 0.15, 0.2, 0.25, 0.3 and 0.35 etc.

[0087] When the distance is the absolute distance of the position or the Euclidean distance of the position vector, the value range of the third preset value is directly related to the application scenario. For example, when the screen is a close-up, the distance between the targets can be compared Therefore, the value range of the third preset value increases accordingly; when the picture is a distant view, the distance between the targets can be relatively small, so that the value range of the third preset value decreases accordingly.

[0088] The value of the third preset value cannot be too large, otherwise the constraint on the position-related information X is too small to be effective, and feature information cannot be introduced at an appropriate time; the value of the third preset value is also It cannot be too small, otherwise the constraint on the location information X is too large, and the feature information calculation will be introduced with a high probability, increasing the calculation cost. Please note that the specific value of the third preset value may be the same as or different from the value of the second preset value, which does not limit the protection scope of the present invention.

[0089] It should be noted that the position of the sensitive object can be automatically output by the detection algorithm, or it can be manually marked, for example: a fixed camera monitoring, the obstacle it faces may be fixed, in this case manually marked Better results.

[0090] In the fourth example, the method further includes: recognizing sensitive objects in each frame of image, and obtaining position-related information of the sensitive objects; the analysis processing includes: calculating the target and the sensitive object in the frame image The overlap area ratio of the object, when the overlap area ratio is greater than the fourth preset value, it is determined that the target is a specific target.

[0091] The inventor discovered through creative work: in some application scenarios, such as in the case of video tracking a multi-person target, if there is a sensitive object (such as an obstacle) that will block the characteristics of the target, then match according to the location-related information X It is likely to go wrong. For this reason, this embodiment first identifies sensitive objects, and then calculates the overlap area ratio between each target and the sensitive object. If the ratio is too large, a certain percentage of the target will be blocked by obstacles. At this time, position-related information X cannot be used for judgment. , That is, a target whose overlap area ratio with a sensitive object is too large is determined as a specific target.

[0092] Since sensitive objects can be identified while recognizing targets, and the calculation of the overlap area ratio is simple and fast, both high accuracy and high efficiency are achieved.

[0093] It should be noted that, for the case where the overlap area ratio is equal to the fourth preset value, either a processing method less than the fourth preset value or a processing method greater than the fourth preset value may be used, both of which are Within the protection scope of the present invention.

[0094] The value range of the fourth preset value may include 0.3-0.8, such as 0.3, 0.4, 0.5, 0.55, 0.6, 0.7, or 0.8. The value of the fourth preset value cannot be too large, otherwise the constraints on the position-related information X are too small, and feature information cannot be introduced for re-identification under appropriate circumstances, and the advantages of the algorithm cannot be highlighted; the fourth preset The value of the value can not be too small, otherwise the constraint on the position-related information X is too large, resulting in two objects that are far apart also need to be identified by feature information calculation, which increases the calculation cost.

[0095] Please note that the specific value of the fourth preset value may be the same as or different from the value of the first preset value, which does not limit the protection scope of the present invention.

[0096] It should be noted that the position of the sensitive object may be automatically output by a detection algorithm, or may be manually marked.

[0097] It should be noted that the present invention can also use any two, any three or even four of the above four examples of processing methods in combination, all of which fall within the protection scope of the present invention, as in the fifth example below.

[0098] The fifth example, which combines the third example and the fourth example, the method further includes: recognizing sensitive objects in each frame of image, and obtaining position-related information of the sensitive objects; the analysis processing includes: calculation The spatial distance between each target in each frame image and the sensitive object in the frame image, the target whose spatial distance is less than the third preset value is judged as a specific target; for all the targets whose spatial distance is greater than the third preset value For the target, calculate the ratio of the overlapping area with the sensitive object in the frame image, and when the ratio of the overlapping area is greater than a fourth preset value, determine that the target is a specific target; set the spatial distance to be greater than the third preset The target whose overlap area ratio is less than the fourth preset value is determined as an unspecified target.

[0099] At this time, it is more in line with the actual application scenario, and the calculation is also simple and faster, so high accuracy and high efficiency are achieved at the same time.

[0100] For the specific processing process of the fifth example, please refer to the specific processing process of the third example and the fourth example, which will not be repeated here.

[0101] At this point, it can be clearly determined that each target in each frame of image is a specific target or a non-specific target (that is, it does not belong to a specific target).

[0102] For a specific target, step S4 is then executed to perform re-identification processing based on the feature information F or simultaneously based on the location-related information and the feature information.

[0103] When the re-identification process is performed only based on the feature information F, the following method can be specifically used: extract the feature vector of the target object through a neural network, and match the feature vector of all (or part) of the tracking target in the previous frame (or previous n frames) Calculate the similarity or distance (Euclidean distance or cosine distance) to obtain the cost matrix. The cost matrix matching method can use the Hungarian algorithm or other similar algorithms. The specific re-identification process is well known to those skilled in the art. No longer.

[0104] When re-identification is performed based on location-related information and feature information at the same time, the cost matrix can be constructed by mixing two parameters with a certain weight λ: X+λF, λ is greater than 0, and the following methods can be specifically used: ..., its specific re-identification The processing procedure is well known to those skilled in the art, and will not be repeated here.

[0105] For non-specific targets, the following steps are performed in sequence: first perform the first re-identification process based on the location-related information, and determine whether the matching is successful, and when the matching fails, perform the second re-identification process based on the characteristic information.

[0106] The first re-identification process can specifically adopt the following method: re-identification is performed by using the spatial position of the detected object and the speed of the position predicted by the tracking result of the first n frames. For example, the specific re-identification process of the SORT algorithm is important to the field The technicians are well-known, so I won't repeat them here.

[0107] The second re-identification processing may specifically adopt the following method: extract the features of the target object by using a neural network, match with all (or part) of the feature vectors of the target in the previous frame (or previous n frames), and calculate the similarity or Distance (Euclidean distance or cosine distance) to get the cost matrix. Using the cost matrix matching method, you can use the Hungarian algorithm or other similar algorithms (for example, some operations in the DeepSORT algorithm). The specific re-identification process is for those skilled in the art It is also well-known, so I won't repeat it here.

[0108] After step S4 is executed, step S6 is successfully matched, and step S7 is executed, step S8 is executed to end.

[0109] It should be noted that the end in step S8 refers to the end of the re-identification process, and other steps after the re-identification process are continued according to the needs of the multi-target tracking method, which are the same as the prior art and will not be repeated here.

[0110] It should be noted that the re-identification processing of non-specific targets can also be realized by other methods. The present invention only limits the judgment and re-identification processing of specific targets, and does not impose any restrictions on the re-identification processing of non-specific targets.

[0111] In the ReID processing of this method, according to the position-related information of the visual image, the use of dynamic adjustment and balance position-related information and characteristic information is realized, so as to realize a high-speed calculation and high-accuracy matching process, which is specifically embodied as follows:

[0112] 1) Realize the dynamic and logical switching of different det in the ReID process, which is mainly reflected in two points: the change of tracking field of view and the position change of det. That is to say, the dynamic adjustment of the ReID logic is automatically realized according to the changes of the scene and the position of the det, and the speed and accuracy of the tracking algorithm are compromised;

[0113] 2) The calculation and matching calculation based on the location-related information X only consumes little, and the new step does not introduce additional calculations;

[0114] 3) According to the location-related information, it can be judged that there is a risk situation, and the appropriate matching calculation classification can be directly selected to improve the accuracy. The jump situation can be adjusted with various preset values, etc., so as to achieve a compromise between speed and accuracy;

[0115] 4) The location information can be judged separately in two situations, or an organic combination of the two:

[0116] The first is the position and speed of det and det;

[0117] The second type, det and global position: For example, the relative distance between det and a fixed obstacle in the image.

[0118] This embodiment also provides a face recognition method, which adopts the above multi-target tracking method for multi-face tracking, which specifically includes the following steps:

[0119] Acquiring a video image to be processed, where the video image to be processed includes a plurality of frame images;

[0120] Performing face recognition processing on each frame image to obtain position-related information and feature information of each face in each frame image, at least one of the frame images includes multiple human faces;

[0121] Analyze and process each face in each frame image separately to determine whether each face belongs to a specific face;

[0122] When the target belongs to a specific face, re-identification processing is performed according to the characteristic information or simultaneously according to the position-related information and the characteristic information.

[0123] For the specific implementation of each step, please refer to the corresponding steps of the above multi-target tracking method, which will not be repeated here.

[0124] This method can simultaneously achieve high accuracy and high efficiency of face recognition.

[0125] This embodiment also provides an intelligent question answering method, which uses the aforementioned face recognition method for user identification, so that the historical information and/or attribute information of the identified user can be directly extracted to combine the historical information and/or attribute of the user Information provides users with answers.

[0126] This method can accurately and efficiently identify users automatically, and combine the historical information and/or attribute information of the identified users when providing answers subsequently, so that the answers are more in line with user needs.

[0127] reference figure 2 As shown, this embodiment also provides a multi-target tracking device, including:

[0128] The image providing module 10 is configured to obtain a video image to be processed, and the video image to be processed includes a plurality of frame images;

[0129] The target recognition processing module 30 is configured to perform target recognition processing on each frame image to obtain position-related information and characteristic information of each target in each frame image, at least one of the frame images includes multiple targets;

[0130] The analysis and processing module 50 is configured to analyze and process each target in each frame image to determine whether each target belongs to a specific target;

[0131] The re-identification processing module 70 is configured to perform re-identification processing according to characteristic information or simultaneously according to location-related information and characteristic information when the target belongs to a specific target.

[0132] Wherein, the analysis and processing module 50 may use position-related information between different targets in the same frame of image to determine whether the current frame belongs to a specific situation; and/or may use position-related information between the same targets in different frames of image Determine whether the target belongs to a specific target.

[0133] As an example, the analysis and processing module 50 may include:

[0134] The overlap area ratio calculation unit is used to calculate the overlap area ratio between two targets in the same frame of image;

[0135] The determining unit is configured to, when the overlapping area ratio is greater than a first preset value, two targets corresponding to the overlapping area ratio greater than the first preset value are specific targets.

[0136] At this time, you can refer to the first example in the above method, which will not be repeated here.

[0137] As another example, the i-1th frame of image includes M targets, and the i-th frame of image includes N targets. The analysis and processing module 50 may include:

[0138] A difference calculation unit, configured to calculate the difference between the position-related information of each target in the i-1 frame image and each target in the i frame image;

[0139] The judging unit is configured to, when the difference between a target in the i-th frame image and the M targets in the i-1th frame image is greater than a second set value, the target is a specific target.

[0140] At this time, you can refer to the second example in the above method, which will not be repeated here.

[0141] In another example, the device further includes: a sensitive object recognition module (not shown in the figure), which is used to recognize sensitive objects in each frame image and obtain position-related information of the sensitive objects; the analysis and processing module 50 can include:

[0142] The spatial distance calculation unit is used to calculate the spatial distance between each target in each frame image and the sensitive object in the frame image;

[0143] The judging unit is configured to judge the target corresponding to the spatial distance less than the third preset value as a specific target.

[0144] At this time, you can refer to the third example in the above method, which will not be repeated here.

[0145] In another example, the device further includes: a sensitive object recognition module (not shown in the figure), which is used to recognize sensitive objects in each frame image and obtain position-related information of the sensitive objects; the analysis and processing module 50 can include:

[0146] The overlap area ratio unit is used to calculate the overlap area ratio of the target and the sensitive object in the frame image;

[0147] The determining unit is configured to determine that the target is a specific target when the overlap area ratio is greater than a fourth preset value.

[0148] At this time, you can refer to the fourth example in the above method, which will not be repeated here.

[0149] As a preferred example, the device further includes: a sensitive object recognition module (not shown in the figure), which is used to recognize sensitive objects in each frame image and obtain position-related information of the sensitive objects; the analysis and processing module 50 can include:

[0150] The spatial distance calculation unit is used to calculate the spatial distance between each target in each frame image and the sensitive object in the frame image;

[0151] An overlapping area ratio calculation unit, configured to calculate an overlapping area ratio of the target whose spatial distance is greater than the third preset value with the sensitive object in the frame image;

[0152] The judging unit is configured to judge the target corresponding to a spatial distance smaller than a third preset value as a specific target, and when the overlap area ratio is greater than a fourth preset value, determine the target as a specific target; A target whose spatial distance is greater than the third preset value and the overlap area ratio is less than the fourth preset value is determined as a non-specific target.

[0153] At this time, you can refer to the fifth example in the above method, which will not be repeated here.

[0154] In addition, the device may further include:

[0155] A location-related information re-identification processing module (not shown in the figure), which is used to perform the first re-identification processing according to the location-related information when the target does not belong to a specific target;

[0156] The matching judgment module (not shown in the figure) is used to judge whether the result of the first re-identification processing is matched successfully;

[0157] The characteristic information processing module (not shown in the figure) is used to perform a second re-identification process according to the characteristic information when the matching is unsuccessful.

[0158] For the specific working process of this device, please refer to the above corresponding method, which will not be repeated here.

[0159] This embodiment also provides a face recognition device, which includes the aforementioned multi-target tracking device.

[0160] This embodiment also provides a robot, at least used for intelligent question answering, which includes the above-mentioned multi-target tracking device.

[0161] This embodiment also provides a computer-readable storage medium on which computer instructions are stored, and when the computer instructions are executed by a processor, the steps of the foregoing method are implemented.

[0162] The computer-readable storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, RAM, Read Only Memory (ROM), erasable Erasable Programmable Read Only Memory (EPROM), flash memory, optical fiber, portable CD-ROM, optical storage device, magnetic storage device, or any suitable combination of the foregoing. In this document, the computer-readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device.

[0163] The computer-readable signal medium may include a data signal propagated in baseband or as a part of a carrier wave, and computer-readable program code is carried therein. This propagated data signal can take many forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the foregoing. The computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium. The computer-readable medium may send, propagate, or transmit the program for use by or in combination with the instruction execution system, apparatus, or device .

[0164] The program code contained on the computer-readable medium can be transmitted by any suitable medium, including, but not limited to, wireless, wire, optical cable, radio frequency (Radio Frequency, RF), etc., or any suitable combination of the above.

[0165] This embodiment also provides a computer device, including a memory and a processor, the memory stores computer instructions that can run on the processor, and the processor executes the above method when the computer instructions are executed. A step of.

[0166] For those skilled in the art, it is obvious that the present invention is not limited to the details of the foregoing exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the foregoing description, and therefore it is intended to fall within the claims. All changes within the meaning and scope of the equivalent elements of are included in the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. Words such as first and second are used to denote names, but do not denote any specific order.

[0167] It should be understood that although an implementation form of the embodiments of the present invention described above may be a computer program product, the methods or devices of the embodiments of the present invention may be implemented by software, hardware, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those of ordinary skill in the art can understand that the above-mentioned methods and devices can be implemented using computer-executable instructions and/or included in processor control codes, for example on a carrier medium such as a disk, CD or DVD-ROM, such as a read-only memory. Such codes are provided on a programmable memory (firmware) or a data carrier such as an optical or electronic signal carrier. The method and device of the present invention can be implemented by hardware circuits such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips and transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc. It can be implemented by software executed by various types of processors, or can be implemented by a combination of the above hardware circuits and software, such as firmware.

[0168] It should be understood that, although several modules or units of the device are mentioned in the above detailed description, this division is only exemplary and not mandatory. In fact, according to the exemplary embodiment of the present invention, the features and functions of two or more modules/units described above can be implemented in one module/unit. Conversely, the features and functions of one module/unit described above Functions can be further divided into multiple modules/units. In addition, some modules/units described above may be omitted in some application scenarios.

[0169] It should be understood that, in order not to obscure the implementation of the present invention, the specification only describes some key, not necessarily necessary technologies and features, and may not describe some features that can be implemented by those skilled in the art.

[0170] Anyone skilled in the art, without departing from the spirit and scope of the present invention, can make various changes and

[0171] Modification, therefore, the protection scope of the present invention should be subject to the scope defined by the claims.