Behavior recognition method and system, storage medium and electronic equipment

Abstract

The invention discloses a behavior recognition method and system, a storage medium and an electronic device, and the method comprises the steps: obtaining first image data and second image data transmitted by a camera device, and audio data, recognizing a target detection object in the second image data, target position information of a target detection object is determined, semantic data corresponding to the audio data are obtained, the target detection object is any one detection object in the second image data, regional image data corresponding to the target position information are intercepted from the first image data, and the target detection object is obtained according to the semantic data and the regional image data. And determining a behavior category of the target detection object, and if the behavior category is an abnormal category, outputting a behavior category report of the target detection object. By adopting the specification, the accuracy and the real-time performance of behavior identification are ensured under the condition of reducing the data processing amount demand.

Description

TECHNICAL FIELD

[0001] The present specification relates to the technical field of computer, and particularly relates to a behavior recognition method and system, a storage medium and an electronic device. BACKGROUND

[0002] Nowadays, with the development of technology, when managing the education scene, in addition to management and guidance by teachers and other personnel, the behavior of objects included in the scene can also be recognized through intelligent recognition function, and how to ensure the efficiency of the management of the education scene according to the intelligent recognition function becomes a problem to be solved. SUMMARY

[0003] The embodiment of the present specification provides a behavior recognition method, system, storage medium and electronic device, which can determine the target position information of the second image data with low resolution, recognize the behavior category of the target detection object in the first image data with high frequency, and further ensure the accuracy and real-time performance of the behavior recognition under the condition of reducing the data processing amount requirement.

[0004] In a first aspect, the embodiment of the present specification provides a behavior recognition method, and the method comprises: obtaining first image data and second image data transmitted by a camera device, and audio data, the resolution of the first image data is higher than the resolution of the second image data; recognizing a target detection object in the second image data, determining target position information of the target detection object, and obtaining semantic data corresponding to the audio data, the target detection object being any detection object in the second image data; cutting region image data corresponding to the target position information in the first image data; determining a behavior category of the target detection object according to the semantic data and the region image data; if the behavior category is an abnormal category, outputting a behavior category report of the target detection object.

[0005] Through the above technical solution, the second image data and the first image data with higher resolution, and the audio data are obtained, after the target position information of the target detection object is determined in the second image data, the region image data is cut in the first image data according to the target position information, the behavior of the target detection object is recognized according to the semantic data corresponding to the audio data and the region image data, and the behavior category of the target detection object is determined, and the behavior category report is generated and output when the behavior category is an abnormal category, so that the target position information determined by the second image data with low resolution is used to recognize the behavior category of the target detection object in the first image data with high frequency, and the accuracy and real-time performance of the behavior recognition are ensured under the condition of reducing the data processing amount requirement.

[0006] With reference to the first aspect and the foregoing implementations, in some possible implementations, the method is applied to a recording and broadcasting device, the recording and broadcasting device is connected with the first computing card and the camera device, the recording and broadcasting device includes a second computing card, a display interface, and a data interface, the camera device is connected with the first computing card and the second computing card, the second computing card is connected with the first computing card through the data interface, and the computing power of the first computing card is higher than the computing power of the second computing card. The second computing card is configured to identify a target detection object in the second image data, determine target position information of the target detection object, and obtain semantic data corresponding to the audio data, and transmit the target position information and the semantic data to the first computing card through the data interface. The first computing card is configured to intercept region image data corresponding to the target position information in the first image data, and determine a behavior category of the target detection object according to the semantic data and the region image data. If the behavior category is an abnormal category, the display interface is controlled to output a behavior category report of the target detection object based on the second computing card.

[0007] According to the foregoing technical solution, the target position information of the target detection object is determined in the second image data with low resolution by the second computing card with low computing power, the region image data is intercepted in the first image data with high resolution by the first computing card with high computing power, and the behavior category of the target detection object is determined according to the semantic data and the region image data, so that the positioning is performed by the computing card with low computing power to reduce the loss of computing power resources, the behavior recognition is performed on the image with high resolution by the computing card with high computing power, and then the efficiency and accuracy of the behavior recognition are improved through the division of the computing cards with high and low computing power.

[0008] With reference to the first aspect and the foregoing implementations, in some possible implementations, the identification of the target detection object in the second image data and the determination of the target position information of the target detection object, and the obtaining of the semantic data corresponding to the audio data, include the following. The detection object set is identified in the second image data according to a pre-trained object recognition model. The target detection object is positioned according to a pre-created positioning coordinate system to determine the target position information of the target detection object, and the target detection object is any detection object in the detection object set. The semantic data corresponding to the audio data is obtained through semantic recognition of the audio data.

[0009] According to the foregoing technical solution, the position information of the target detection object is determined in the second image data with low resolution, and the semantic data of the audio data is obtained, so that the data calculation amount required for subsequent behavior recognition of the target detection object is reduced, and the accuracy of subsequent recognition of the behavior category of the target detection object is improved according to the semantic data.

[0010] With reference to the first aspect and the above implementation manners, in some possible implementation manners, the intercepting, from the first image data, the region image data corresponding to the target position information comprises: determining a pixel ratio between the first image data and the second image data; adjusting the target position information based on the pixel ratio to obtain intercept position information; intercepting the first image data based on the intercept position information to obtain the region image data.

[0011] According to the above technical solution, the target position information is adjusted based on the pixel ratio between the first image data and the second image data to obtain intercept position information, and then the region image data is intercepted from the first image data based on the intercept position information, so that the region image data including the target detection object is accurately obtained from the first image data.

[0012] With reference to the first aspect and the above implementation manners, in some possible implementation manners, the behavior category of the target detection object is determined based on the semantic data and the region image data, comprising: obtaining first historical image data corresponding to the target detection object within a preset time length before a current time, the current time being a time when the first image data is obtained; sorting the first historical image data and the region image data according to the current time and historical times of the first historical image data to obtain a first detection data set arranged in time sequence; determining the behavior category of the target detection object based on the semantic data and the first detection data set.

[0013] According to the above technical solution, the first detection data set arranged in time sequence is obtained based on the region image data and the first historical image data corresponding to the target detection object, and the behavior recognition is performed in combination with the semantic data, so that the accuracy of the determined behavior category is improved according to the continuous action and the semantic data.

[0014] With reference to the first aspect and the above implementation manners, in some possible implementation manners, the first historical image data corresponding to the target detection object within a preset time length before a current time is obtained, comprising: obtaining second historical image data within a preset time length before the current time, the resolution of the second historical image data being the same as that of the first image data; intercepting the second historical image data based on the target detection object to obtain the first historical image data corresponding to the target detection object.

[0015] With reference to the first aspect and the above implementation manners, in some possible implementation manners, the behavior category of the target detection object is determined according to the semantic data and the first detection data set, including: scaling data sizes of the image data in the first detection data set to a target size to obtain a second detection data set; determining the behavior category of the target detection object according to the semantic data and the second detection data set.

[0016] With reference to the first aspect and the above implementation manners, in some possible implementation manners, the behavior category of the target detection object is determined according to the semantic data and the second detection data set, including: determining an action feature corresponding to the second detection data set according to a time sequence code of the image data in the second detection data set; obtaining a semantic feature of the semantic data, the semantic feature and the action feature having the same data dimension; determining an action behavior of the target detection object according to the action feature and the semantic feature, and identifying the action behavior to determine the behavior category of the target detection object.

[0017] With reference to the first aspect and the above implementation manners, in some possible implementation manners, if the behavior category is an abnormal category, a behavior category report of the target detection object is output, including: matching the behavior category with each abnormal behavior in a preset abnormal behavior set; if the abnormal behavior set includes an abnormal behavior matching the behavior category, determining that the behavior category is an abnormal category, and generating and outputting a behavior category report of the target detection object based on the target location information and the behavior category.

[0018] By the above technical solution, the behavior category of the target detection object is determined, and when the behavior category is an abnormal behavior, a behavior category report is output, so that the abnormal behavior can be prompted in time, and the efficiency of discovering the abnormal behavior is ensured.

[0019] With reference to the first aspect and the above implementation manners, in some possible implementation manners, the behavior category report of the target detection object is generated and output based on the target location information and the behavior category, including: determining a current time at which the first image data is obtained, generating a behavior category report of the target detection object based on the target location information, the behavior category and the current time, the behavior category report being used to describe the behavior category of the target detection object in the target location information; outputting the behavior category report on a display interface of the recording and broadcasting device.

[0020] According to the technical solution, the behavior of the target detection object at the current moment is reported according to the target position information and the behavior category, so that the target detection object making an abnormal behavior can be prompted in time, and the efficiency and timeliness of finding the abnormal behavior are ensured.

[0021] In a second aspect, an embodiment of the present specification provides a behavior recognition device, the device comprising: a data acquisition unit configured to acquire first image data and second image data transmitted by a camera device, and audio data, the resolution of the first image data being higher than the resolution of the second image data; a position information determination unit configured to identify a target detection object in the second image data, determine target position information of the target detection object, and acquire semantic data corresponding to the audio data, the target detection object being any one detection object in the second image data; an image determination unit configured to intercept region image data corresponding to the target position information in the first image data; a behavior recognition unit configured to determine a behavior category of the target detection object according to the semantic data and the region image data; a reporting unit configured to output a behavior category report of the target detection object if the behavior category is an abnormal category.

[0022] In a third aspect, an embodiment of the present specification provides a behavior recognition system, characterized in that the system comprises a recording and broadcasting device, a first computing power card and a camera device, the recording and broadcasting device comprising a second computing power card, a display interface and a data interface, the camera device being connected to the first computing power card and the second computing power card, the second computing power card being connected to the first computing power card through the data interface, the computing power of the first computing power card being higher than the computing power of the second computing power card; The first computing power card acquires first image data transmitted by the camera device, and the second computing power card acquires second image data and audio data transmitted by the camera device, the resolution of the first image data being higher than the resolution of the second image data; The second computing power card identifies a target detection object in the second image data, determines target position information of the target detection object, and acquires semantic data corresponding to the audio data, transmits the target position information and the semantic data to the first computing power card through the data interface, and the target detection object is any one detection object in the second image data; The first computing power card intercepts region image data corresponding to the target position information in the first image data; The first computing power card determines a behavior category of the target detection object according to the semantic data and the region image data, and transmits the behavior category to the second computing power card through the data interface; If the behavior category is an abnormal category, the second computing power card controls the display interface to output a behavior category report of the target detection object.

[0023] In a fourth aspect, an embodiment of the present specification provides a computer storage medium, the computer storage medium storing a plurality of instructions, the instructions being suitable for being loaded by a processor and executing steps of the method described above.

[0024] In a fifth aspect, an embodiment of the present specification provides an electronic device, comprising: a processor and a memory; wherein the memory stores a computer program, the computer program being suitable for being loaded by the processor and executing steps of the method described above. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the specification or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the specification, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0026] Figure 1 A system architecture diagram of a behavior recognition method provided by an embodiment of the present specification; Figure 2 A system architecture diagram of a behavior recognition method provided by an embodiment of the present specification; Figure 3 A flowchart of a behavior recognition method provided by an embodiment of the present specification; Figure 4 A flowchart of a behavior recognition method provided by an embodiment of the present specification; Figure 5 An example schematic diagram of target position information provided by an embodiment of the present specification; Figure 6 An example schematic diagram of determining area image data provided by an embodiment of the present specification; Figure 7 An example schematic diagram of a first detection data set provided by an embodiment of the present specification; Figure 8 A structural schematic diagram of a behavior recognition device provided by an embodiment of the present specification; Figure 9 A structural schematic diagram of an electronic device provided by an embodiment of the present specification. DETAILED DESCRIPTION

[0027] To make the features and advantages of this specification more apparent and understandable, the technical solutions in this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0028] Please see Figure 1 The diagram below illustrates a system architecture for a behavior recognition method as provided in the embodiments of this specification. Figure 1 As shown in the embodiments of this specification, the behavior recognition method can be applied to recording and broadcasting equipment to realize the process of reporting target detection objects that perform abnormal behaviors. The system structure provided in the embodiments of this specification mainly includes a recording and broadcasting equipment 10, a first computing power card 20, and a camera device 30. The recording and broadcasting equipment 10 can be a terminal device with data processing capabilities, such as a smartphone, personal computer, or interactive flat panel. The recording and broadcasting equipment 10 includes a second computing power card 101, a display interface 102, and a data interface 103. The camera device 30 is connected to the first computing power card 20 and the second computing power card 101, and the second computing power card 102 is connected to the first computing power card 20 through the data interface 103. The second computing power card 101 can be a processing module with data processing capabilities and is located inside the recording and broadcasting equipment 10. The display interface 102 can be a visualization module in the recording and broadcasting equipment 10, such as the display screen of the recording and broadcasting equipment 10. The data interface 103 can be a module for connecting to external devices. The first computing power card 20 can be a data processing module, and its computing power is higher than that of the second computing power card 101. The camera device 30 can be a device set up in the detection scene, with image capture and audio recording functions. It can be understood that in order to acquire image data of different resolutions at the same time, the camera device 30 is a device that supports dual bitstreams.

[0029] Furthermore, such as Figure 2 As shown, Figure 2 The first computing card 20 can also be installed in the recording and broadcasting equipment 10 as a built-in computing card. The first computing card 20 and the second computing card 101 are directly connected. The specific configuration of the first computing card 20 can be configured according to the actual situation. Alternatively, the first computing card 20 can also be connected to the second computing card 101 through the data interface 103, which can be configured according to the actual situation.

[0030] In the related art, in the process of behavior recognition in an education scene, the method used is to recognize the collected single-frame data, but the determined recognition result is limited by the reasons of less data content and insufficient device computing power, resulting in the problem of insufficient accuracy of the determined recognition result, which cannot meet the application requirements of behavior recognition.

[0031] In the embodiments of the present specification, the first computing power card 20 acquires the first image data transmitted by the camera device 30, the second computing power card 101 acquires the second image data and the audio data transmitted by the camera device 30, the second computing power card 101 recognizes the target detection object in the second image data, determines the target position information of the target detection object, and acquires the semantic data corresponding to the audio data, and transmits the target position information and the semantic data to the first computing power card 20 through the data interface. The first computing power card 20 intercepts the region image data corresponding to the target position information in the first image data, and then determines the behavior category of the target detection object according to the semantic data and the region image data, and transmits the behavior category to the second computing power card 101 through the data interface. If the behavior category is an abnormal category, the second computing power card 101 controls the display interface 102 of the recording and broadcasting device 10 to output the behavior category report of the target detection object. Thus, the target position information determined by the low-resolution second image data is used to recognize the behavior category of the target detection object in the high-frequency first image data, thereby ensuring the accuracy and real-time performance of the behavior recognition while reducing the data processing amount requirement.

[0032] Based on the system architecture shown in FIG. 1, the behavior recognition method provided by the embodiments of the present specification will be described in detail below. Figure 1 Figure 3 The behavior recognition method provided by the embodiments of the present specification will be described in detail below.

[0033] Please refer to FIG. 2, Figure 3 FIG. 2 shows a flowchart of a behavior recognition method provided by the embodiments of the present specification. As shown in FIG. 2, the method can include the following steps S101-S105. Figure 3

[0034] S101, acquiring first image data and second image data transmitted by a camera device, and audio data; ​​In one embodiment, the first image data and the second image data photographed by the camera device for detecting the scene are acquired, and the audio data recorded by the camera device when the first image data and the second image data are photographed are acquired. The camera device can be a device with image photographing function and audio recording function arranged in the detection scene, and the detection scene can be a scene in which behavior detection is required, for example, a classroom, an examination room, etc. The first image data can be data in the main stream of the camera device, and the second image data can be data in the auxiliary stream of the camera device. The resolution of the first image data is higher than that of the second image data, for example, the resolution of the first image data can be 7680*4320, and the resolution of the second image data can be 1280*720, etc. The audio data can be audio data collected by the camera device when the first image data and the second image data are collected.

[0035] In one embodiment, the target detection object in the second image data is identified by using the pre-trained object recognition model, the target position information of the target detection object in the second image data is determined, and the semantic data corresponding to the audio data is acquired. In one embodiment, the target detection object in the second image data is identified by using the pre-trained object recognition model, the target position information of the target detection object in the second image data is determined, and the semantic data corresponding to the audio data is acquired. The object recognition model can be a model with the function of identifying a person or an object. The target detection object can be any detection object in the second image data, and the detection object can be an object in the detection scene that needs to be recognized, for example, a student and a teacher in a classroom, a test taker and a proctor in an examination room, etc. The positioning coordinate system can be a coordinate system used for positioning each pixel point in the image data, which can be a plane rectangular coordinate system, etc. The target position information can be data used for marking the position of the target detection object, which represents the area in which the target detection object is located in the second image data, and can be four vertex coordinates of a frame used for marking the target detection object, etc. The audio data is subjected to speech recognition, and the semantic data corresponding to the audio data is acquired. The semantic data can be text data representing the verbal communication in the detection scene.

[0036] In one embodiment, the target detection object in the second image data is identified by using the pre-trained object recognition model, the target position information of the target detection object in the second image data is determined, and the semantic data corresponding to the audio data is acquired. In one embodiment, the target detection object in the second image data is identified by using the pre-trained object recognition model, the target position information of the target detection object in the second image data is determined, and the semantic data corresponding to the audio data is acquired.

[0037] It can be understood that, since the first image data and the second image data include the same range of detected scenes, the target detection object can be accurately positioned in the first image data according to the target position information determined in the second image data, so that the process of object recognition on the first image data is saved, and the calculation amount of full-image recognition on the high-resolution first image data is reduced.

[0038] In S104, a behavior category of the target detection object is determined according to the semantic data and the region image data. In one embodiment, the behavior of the target detection object is recognized according to the semantic data and the region image data by cross-attention combination, and the behavior category of the target detection object is determined. The behavior category can be a category representing the behavior performed by the target detection object, for example, can be hand raising, playing a mobile phone, etc.

[0039] Specifically, when the behavior is recognized according to the semantic data and the region image data, the target part of the target detection object that needs to be focused can be determined according to the semantic data, and the action performed by the target part is focused. For example, the semantic data is "who can answer this question, raise your hand to answer", the target part can be the hand, and the possible action can be hand raising, etc.

[0040] In S105, if the behavior category is an abnormal category, a behavior category report of the target detection object is output. In one embodiment, the determined behavior category is judged to determine whether the behavior category is an abnormal category, if yes, a behavior category report of the target detection object is generated according to the target position information and the behavior category, and the behavior category report is output on the display interface of the recording and broadcasting device. The behavior category report can be report data used for describing the behavior category of the target detection object, and the data type of the behavior category report can be text data, or a combination of text data and image data, which can be set according to actual conditions.

[0041] In the embodiments of the present specification, by acquiring the second image data and the first image data with higher resolution, and the audio data, after the target position information of the target detection object is determined in the second image data, the region image data is intercepted in the first image data according to the target position information, the behavior of the target detection object is recognized according to the semantic data corresponding to the region image data and the audio data, and the behavior category of the target detection object is determined, and the behavior category report is generated and output when the behavior category is an abnormal category, so that the behavior category of the target detection object is recognized in the high-frequency first image data according to the target position information determined by the low-resolution second image data, and the accuracy and real-time performance of the behavior recognition are ensured while reducing the data processing amount.

[0042] Please refer toFigure 4 A flowchart of a behavior recognition method is provided for the embodiments of the present specification. As shown, it can include the following steps S201-S208. Figure 4

[0043] S201, acquiring first image data and second image data transmitted by a camera device, and audio data; In an embodiment, the first image data and the second image data shot by the camera device for detecting the scene are acquired, and the audio data recorded by the camera device when shooting the first image data and the second image data is acquired. The camera device can be a device with image shooting function and audio recording function arranged in the detection scene, and the detection scene can be a scene that needs to be detected, such as a classroom, an examination room, etc. The first image data can be data in the main stream of the camera device, and the second image data can be data in the auxiliary stream of the camera device. The resolution of the first image data is higher than that of the second image data, for example, the resolution of the first image data can be 7680*4320, and the resolution of the second image data can be 1280*720, etc. The audio data can be audio data collected by the camera device when collecting the first image data and the second image data.

[0044] S202, recognizing a target detection object in the second image data, determining target position information of the target detection object, and acquiring semantic data corresponding to the audio data; In an embodiment, the second image data is subjected to portrait recognition, the target detection object is determined in the second image data, and the target position information of the target detection object in the second image data is determined. The target detection object is any detection object in the second image data, which can be an object in the detection scene that needs to be recognized, such as students and teachers in a classroom, examinees and invigilators in an examination room, etc. The target position information can be data for marking the position of the target detection object, representing the area where the target detection object is located in the second image data, which can be four vertex coordinates of a frame for marking the target detection object. The audio data is subjected to voice recognition, and the semantic data corresponding to the audio data is acquired. The semantic data can be text data representing the verbal communication in the detection scene.

[0045] ​Specifically, the manner of determining the target position information can be: determining a set of detected objects in the second image data according to a pre-trained object recognition model, positioning a target detected object according to a pre-created positioning coordinate system, and determining target position information of the target detected object, the target detected object being any detected object in the set of detected objects. The object recognition model can be a model having the function of recognizing portraits or objects. The set of detected objects can be a set of detected objects detected in the second image data. The positioning coordinate system can be a coordinate system used for positioning each pixel point in the image data, and can be a plane rectangular coordinate system, etc. The target detected object can be determined by portrait recognition, and the corresponding target position information can be determined according to the size of the target detected object in the second image data.

[0046] As shown in the examples, Figure 5 Figure 5 The set of detected objects in the image data includes four detected objects, and the position of the determined target detected object "A" is determined according to the positioning coordinate system, and the target position information "M(x1, y1)", "N(x2, y2)", "O(x3, y3)", and "P(x4, y4)" corresponding to the target detected object "A" are determined.

[0047] It should be noted that the target position information of the target detected object and the semantic data corresponding to the audio data can be obtained according to the second computing card, and the second computing card transmits the target position information and the semantic data to the first computing card through a data interface. The second computing card can be a data processing module arranged in the recording and broadcasting device. The recording and broadcasting device can be a device for processing data transmitted by a camera device to detect and recognize detected objects in a detection scene. The first computing card can be a data processing module having a data processing function, and the computing power of the first computing card is higher than that of the second computing card. The first computing card can be connected to the second computing card through a data interface in the recording and broadcasting device. The first computing card can be arranged outside or inside the recording and broadcasting device, and the specific arrangement manner of the first computing card can be set according to actual conditions. The data interface can be a module for connecting the first computing card and the second computing card in the recording and broadcasting device.

[0048] S203, intercepting region image data corresponding to the target position information in the first image data; In one embodiment, the target detected object is positioned in the first image data according to the target position information, and the region corresponding to the target position information in the first image data is intercepted as region image data. The region image data can be image data obtained by intercepting the region where the target detected object is located in the first image data.

[0049] ​Specifically, the manner of intercepting the region image data can be: determining a pixel ratio between the first image data and the second image data, adjusting the target position information based on the pixel ratio to obtain intercept position information, and intercepting the first image data based on the intercept position information to obtain the region image data. The pixel ratio can be a ratio representing the resolution between the first image data and the second image data. For example, the resolution of the first image data is 7680*4320, and the resolution of the second image data is 1280*720. The pixel ratio is 6. The intercept position information can be obtained by scaling the coordinates in the target position information by the pixel ratio.

[0050] It can be understood that, since the first image data and the second image data include the same detection scene range, the target detection object can be accurately positioned in the first image data according to the target position information determined in the second image data, thereby saving the process of object recognition on the first image data and reducing the calculation amount required for full-image recognition on the high-resolution first image data.

[0051] For example, as shown in FIG. 2, the target detection object is a person, and the first image data and the second image data include the same detection scene range. Figure 6 Figure 6 According to the pixel ratio, the target position information determined in the second image data is adjusted to obtain intercept position information, and the region image data corresponding to the target detection object is obtained by intercepting the first image data based on the intercept position information.

[0052] S204, obtaining first historical image data corresponding to the target detection object within a preset time length before the current time; In one embodiment, the second historical image data within the preset time length before the current time is obtained, the resolution of the second historical image data is the same as the resolution of the first image data, and the first historical image data corresponding to the target detection object is obtained by intercepting the second historical image data. The current time can be the time when the first image data is obtained by the camera device, for example, it can be 2025 / 01 / 01 09:00:10, etc. The preset time length can be a time length set in advance for obtaining historical data. The specific value of the preset time length can be set according to actual conditions, for example, it can be 3 seconds, or ten frames, etc. The second historical image data can be image data captured by the camera device within the preset time length and for the detection scene. The first historical image data can be image data including only the target detection object intercepted from the second historical image data.

[0053] ​It should be noted that, since the recording and broadcasting device will intercept the region image data corresponding to the target detection object in the first image data according to the second image data after obtaining the first image data and the second image data, each second historical image data has the first historical image data corresponding to the target detection object, which can be directly searched according to the target detection object to obtain the first historical image data corresponding to the target detection object.

[0054] S205, sorting each first historical image data and region image data according to the current time and the historical time of each first historical image data, to obtain a first detection data set arranged according to time sequence; In one embodiment, the historical time of each first historical image data is determined, and each first historical image data and region image data is sorted according to the time sequence of the current time and the historical time, to obtain a first detection data set arranged according to time sequence. The historical time can be the time when each first historical image data is obtained. The first detection data set can be a data set for representing the action behavior of the target detection object performed within a period of time before the current time and a preset time length.

[0055] Among them, the time sequence can be arranged from the historical time with the longest interval time length to the current time according to the interval time length of each historical time and the current time. For example, as shown in Figure 7 Figure 7 The first data set is obtained according to the historical time of each first historical image data and the current time.

[0056] It should be noted that, if the number of first historical image data does not reach the preset number, the first detection data set is not generated for behavior recognition, so as to improve the accuracy of the obtained behavior recognition. It can be understood that the first detection data set can also be generated according to the obtained first historical image data and region image data, or only the region image data is used for behavior recognition. The specific setting can be made according to the actual situation.

[0057] S206, determining the behavior category of the target detection object according to the semantic data and the first detection data set; ​In an embodiment, the data size of each image data in the first detection data set is scaled to a target size to obtain a second detection data set, and a behavior category of the target detection object is determined according to the semantic data and the second detection data set. The data size can be the resolution of each image data in the first detection data, which can be 192*192, for example. The target size can be a pre-set resolution, which can be 448*448, for example. The second detection data set can be a data set whose image data size is the target size. The behavior category can be a category representing the action behavior performed by the target detection object, which can be “raising hands”, “playing a mobile phone”, and the like.

[0058] Specifically, the manner of determining the behavior category of the target detection object can be: determining an action feature corresponding to the second detection data set according to the time sequence coding of each image data in the second detection data set. The time sequence coding can be coding representing the time sequence of each image data. The action feature can be a feature obtained by extracting the action of the target detection object in the second detection data set according to a pre-trained visual encoder of a behavior recognition model, for example, determining the action continuity of the target detection object according to the part change of the target detection object, and the like. The behavior recognition model can be a model pre-trained in the first computing card, having modules such as visual data processing, text data processing, and classifier, so as to be able to recognize the behavior category of the target detection object.

[0059] The semantic feature of the semantic data is obtained, and the data dimensions of the semantic feature and the action feature are the same. The manner of obtaining the semantic feature can be a feature extracted by a text encoder in the pre-trained behavior recognition model.

[0060] The action behavior of the target detection object is determined according to the action feature and the semantic feature, and the action behavior is recognized to determine the behavior category of the target detection object. Among them, the manner of determining the action behavior can be to recognize the semantic feature and the action feature by cross-attention according to the behavior recognition model, to improve the visual feature part related to the semantic feature of the behavior recognition model, for example, the semantic feature is “who can answer this question, raise your hand to answer”, and the behavior recognition model is biased to recognize the part of “hand” in the visual feature, thereby improving the accuracy of determining the action behavior of the target detection object. The action behavior can be action information indicating the action performed by the target detection object, such as the action of placing the arm and palm on the desktop to raising, and the like. Then, the action behavior is classified by the classifier in the behavior recognition model to determine the behavior category corresponding to the target detection object.

[0061] It should be noted that steps S203 to S206 are executed by the first computing card, and after the first computing card determines the behavior category of the target detection object, the behavior category is transmitted to the second computing card through the data interface.

[0062] S207, match the behavior category with each abnormal behavior in the preset abnormal behavior set; S208, if the abnormal behavior set includes the abnormal behavior matching the behavior category, determine the behavior category as an abnormal category, and generate and output a behavior category report of the target detection object based on the target position information and the behavior category; In one embodiment, the determined behavior category is matched with each abnormal behavior in the preset abnormal behavior set respectively. The abnormal behavior set can be a behavior set including at least one behavior labeled as an abnormal behavior. The abnormal behavior can be a behavior category indicating that a prompt needs to be given in the detection scene. According to different detection scenes, the abnormal behaviors in the abnormal behavior set are different. For example, if the detection scene is a classroom, the abnormal behavior set can include raising hands, playing mobile phones, etc.; if the detection scene is an examination room, the abnormal behavior set can include whispering, passing notes, etc. The abnormal category can be used to indicate that the behavior category of the target detection object is abnormal and a prompt is needed. Based on the target position information and the behavior category, a behavior category report describing the behavior category of the target detection object is generated. The behavior category report can be report data for describing the behavior category of the target detection object. The data type of the behavior category report can be text data, or a combination of text data and image data, which can be set according to actual conditions. After generating the behavior category report, the behavior category report is output on the display interface of the recording and broadcasting device to prompt the behavior category of the target detection object. It can be understood that, in order to facilitate the description of the target detection object, the target position information is combined to facilitate the user to determine the specific position of the target detection object, thereby improving the convenience of the user viewing the behavior category report.

[0063] Specifically, the behavior category report can be generated in the following manner: determining a current time when the first image data is obtained, generating a behavior category report of the target detection object based on the target position information, the behavior category, and the current time, the behavior category report being used to describe the behavior category of the target detection object in the target position information, and outputting the behavior category report on the display interface of the recording and broadcasting device. It should be noted that the target position information in the behavior category report can be the coordinates of the target detection object, or can be converted into text that is easy to understand according to the detection scene, such as "the first row and the first column near the podium and the window".

[0064] For example, if the behavior category report only includes text data, the behavior category report can be "a student located in the first row and the first column near the podium and the window in the image is playing a mobile phone". If the behavior category report includes image data and text data, the behavior category report can be the first image data with an image frame generated according to the target position information and the text data. The specific setting can be made according to actual conditions.

[0065] It should be noted that the matching of the behavior category and the generation of the behavior category report can be performed by the second computing power card. After the behavior category report is generated, the second computing power card controls the display interface of the recording and broadcasting device to output the behavior category report. The display interface can be a display module in the recording and broadcasting device, which has a data display function, and can be a display screen of the recording and broadcasting device, etc.

[0066] It can be understood that if the behavior category does not match each abnormal behavior in the abnormal behavior set, it can be considered that the behavior category of the target detection object is not abnormal, and the behavior of the target detection object does not need to be prompted, and therefore the behavior category report does not need to be generated for the current moment.

[0067] In the embodiments of the present specification, by obtaining the second image data and the first image data with higher resolution, and the audio data, after determining the target position information of the target detection object in the second image data, the region image data is intercepted in the first image data according to the target position information, the behavior of the target detection object is identified according to the semantic data corresponding to the region image data and the audio data, and the behavior category of the target detection object is determined, and the behavior category report is generated and output when the behavior category is an abnormal category, so that the target position information determined by the low-resolution second image data is used to identify the behavior category of the target detection object in the high-frequency first image data, and the accuracy and real-time performance of the behavior identification are ensured while reducing the data processing amount. Further, the region image data and the first historical image data are generated to generate the first detection data set according to the current moment and the preset time length, so as to identify the behavior of the target detection object within a certain period of time, thereby improving the accuracy of the behavior category of the target detection object.

[0068] Based on the system architecture shown in Figure 1 , the behavior identification device provided by the embodiments of the present specification will be described in detail below. It should be noted that Figure 8 , the behavior identification device provided by the embodiments of the present specification will be described in detail below. It should be noted that Figure 8 , the behavior identification device in the embodiments of the present specification is configured to execute the method of the embodiments of the present specification Figures 2 to 7 , the method of the embodiments of the present specification, only the parts related to the embodiments of the present specification are shown, and the specific technical details not disclosed are described with reference to the embodiments of the present specification Figures 2 to 7 .

[0069] Please refer to Figure 8 , the behavior identification device provided by the embodiments of the present specification will be described in detail below. It should be noted that Figure 8 , the behavior identification device 1 of the embodiments of the present specification can include a data acquisition unit 11, a position information determination unit 12, an image determination unit 13, a behavior identification unit 14, and a report unit 15.

[0070] The data acquisition unit 11 is configured to acquire first image data and second image data transmitted by the camera device, and audio data, the resolution of the first image data being higher than that of the second image data; The position information determination unit 12 is configured to identify a target detection object in the second image data, determine target position information of the target detection object, and acquire semantic data corresponding to the audio data, the target detection object being any one detection object in the second image data; The image determination unit 13 is configured to intercept region image data corresponding to the target position information in the first image data; The behavior recognition unit 14 is configured to determine a behavior category of the target detection object according to the semantic data and the region image data; The reporting unit 15 is configured to output a behavior category report of the target detection object if the behavior category is an abnormal category.

[0071] Optionally, the position information determination unit 12 is further configured to: identify a detection object set in the second image data according to a pre-trained object recognition model; position the target detection object according to a pre-created positioning coordinate system to determine target position information of the target detection object, the target detection object being any one detection object in the detection object set; perform semantic recognition on the audio data to acquire semantic data corresponding to the audio data.

[0072] Optionally, the image determination unit 13 is further configured to: determine a pixel ratio between the first image data and the second image data; adjust the target position information based on the pixel ratio to obtain interception position information; intercept the first image data based on the interception position information to obtain the region image data.

[0073] Optionally, the behavior recognition unit 14 is further configured to: acquire first historical image data corresponding to the target detection object within a preset time length before a current time, the current time being a time when the first image data is acquired; sort the first historical image data and the region image data according to a time sequence to obtain a first detection data set arranged according to a time sequence, according to the current time and historical times of the first historical image data; determine the behavior category of the target detection object according to the semantic data and the first detection data set.

[0074] Optionally, the behavior recognition unit 14 is further configured to: acquire second historical image data in a preset time period before the current time, the resolution of the second historical image data being the same as the resolution of the first image data; perform intercepting on the second historical image data based on the target detection object, and acquire first historical image data corresponding to the target detection object.

[0075] Optionally, the behavior recognition unit 14 is further configured to: scale the data size of each image data in the first detection data set to a target size to obtain a second detection data set; determine the behavior category of the target detection object according to the semantic data and the second detection data set.

[0076] Optionally, the behavior recognition unit 14 is further configured to: determine the action feature corresponding to the second detection data set according to the time sequence coding of each image data in the second detection data set; acquire a semantic feature of the semantic data, the data dimension of the semantic feature being the same as that of the action feature; determine the action behavior of the target detection object according to the action feature and the semantic feature, and recognize the action behavior to determine the behavior category of the target detection object.

[0077] Optionally, the report unit 15 is further configured to: match the behavior category with each abnormal behavior in a preset abnormal behavior set; if the abnormal behavior set includes an abnormal behavior matching the behavior category, determine that the behavior category is an abnormal category, and generate and output a behavior category report of the target detection object based on the target location information and the behavior category.

[0078] Optionally, the report unit 15 is further configured to: if there are multiple target speech scores greater than the score threshold, acquire the historical speech time of the candidate person corresponding to the target speech score in the second historical detection time period; determine the speaker among the candidate persons based on the historical speech time.

[0079] In the embodiment of the present application, by acquiring the second image data and the first image data with higher resolution, and audio data, after determining the target position information of the target detection object in the second image data, the region image data is intercepted in the first image data according to the target position information, the behavior of the target detection object is identified according to the semantic data corresponding to the region image data and the audio data, and the behavior category of the target detection object is determined, and the behavior category report is generated and output when the behavior category is an abnormal category, so that the target position information determined by the second image data with low resolution is used to identify the behavior category of the target detection object in the first image data with high frequency, and the accuracy and real-time performance of the behavior identification are ensured while reducing the data processing amount. Further, the region image data and the first historical image data are generated to generate the first detection data set according to the current time and the preset time length, so as to identify the behavior of the target detection object in a certain period of time, thereby improving the accuracy of the behavior category of the target detection object.

[0080] The embodiment of the present application also provides a computer storage medium, which can store a plurality of program instructions, the program instructions being suitable for being loaded and executed by a processor to perform the method steps of the embodiment as shown in the above Figures 1-7 The specific implementation process can refer to the specific description of the embodiment as shown in the above Figures 1-7 The specific implementation process can refer to the specific description of the embodiment as shown in the above

[0081] The embodiment of the present application also provides a computer program product, which stores at least one instruction, and the at least one instruction is loaded and executed by the processor to perform the behavior identification method of the embodiment as shown in the above Figures 1-7 The specific implementation process can refer to the specific description of the embodiment as shown in the above Figures 1-7 The specific implementation process can refer to the specific description of the embodiment as shown in the above

[0082] Referring to Figure 9 , a structural schematic diagram of an electronic device is provided for the embodiment of the present application. As shown in the above Figure 9 , the electronic device 1000 can include at least one processor 1001, such as a CPU, at least one network interface 1004, an input and output interface 1003, a memory 1005, and at least one communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the components. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk memory. As shown in the above Figure 9As shown, the memory 1005 as a computer storage medium can include an operating system, a network communication module, an input / output interface module, and a behavior recognition application.

[0083] In Figure 9 In the electronic device 1000 shown, the input / output interface 1003 is mainly used to provide an interface for user input and obtain data input by the user.

[0084] In one embodiment, the processor 1001 can be used to call the behavior recognition application stored in the memory 1005, and specifically perform the following operations: Obtain the first image data and the second image data transmitted by the camera device, and the audio data, the resolution of the first image data being higher than that of the second image data; Identify the target detection object in the second image data, determine the target position information of the target detection object, and obtain the semantic data corresponding to the audio data, the target detection object being any detection object in the second image data; Crop the region image data corresponding to the target position information in the first image data; According to the semantic data and the region image data, determine the behavior category of the target detection object; If the behavior category is an abnormal category, output the behavior category report of the target detection object.

[0085] Optionally, the method is applied to a recording and broadcasting device, the recording and broadcasting device is connected with the first computing power card and the camera device, the recording and broadcasting device includes a second computing power card, a display interface, and a data interface, the camera device is connected with the first computing power card and the second computing power card, the second computing power card is connected with the first computing power card through the data interface, and the computing power of the first computing power card is higher than that of the second computing power card; The second computing power card is used to identify the target detection object in the second image data, determine the target position information of the target detection object, and obtain the semantic data corresponding to the audio data, and transmit the target position information and the semantic data to the first computing power card through the data interface; The first computing power card is used to crop the region image data corresponding to the target position information in the first image data, and determine the behavior category of the target detection object according to the semantic data and the region image data; If the behavior category is an abnormal category, the display interface is controlled to output the behavior category report of the target detection object based on the second computing power card.

[0086] Optionally, when the processor 1001 identifies the target detection object in the second image data, determines the target position information of the target detection object, and obtains the semantic data corresponding to the audio data, the processor 1001 specifically performs the following operations: The object recognition model is pre-trained, and the detection object set is recognized in the second image data; The target detection object is positioned according to the positioning coordinate system, and target position information of the target detection object is determined, the target detection object being any detection object in the detection object set; The audio data is subjected to semantic recognition to obtain semantic data corresponding to the audio data.

[0087] Optionally, when the processor 1001 executes the region image data corresponding to the target position information in the first image data, the following operations are specifically performed: The pixel ratio between the first image data and the second image data is determined; The target position information is adjusted based on the pixel ratio to obtain the interception position information; The first image data is intercepted based on the interception position information to obtain the region image data.

[0088] Optionally, when the processor 1001 executes the behavior category of the target detection object according to the semantic data and the region image data, the following operations are specifically performed: The first historical image data corresponding to the target detection object in a preset time period before the current time is obtained, the current time being the time when the first image data is obtained; The first historical image data and the region image data are sorted according to the current time and the historical time of each first historical image data to obtain a first detection data set arranged in time sequence; The behavior category of the target detection object is determined according to the semantic data and the first detection data set.

[0089] Optionally, when the processor 1001 executes the first historical image data corresponding to the target detection object in a preset time period before the current time, the following operations are specifically performed: The second historical image data in a preset time period before the current time is obtained, the resolution of the second historical image data being the same as that of the first image data; The second historical image data is intercepted based on the target detection object to obtain the first historical image data corresponding to the target detection object.

[0090] Optionally, when the processor 1001 executes the behavior category of the target detection object according to the semantic data and the first detection data set, the following operations are specifically performed: The data size of each image data in the first detection data set is scaled to a target size to obtain a second detection data set; The behavior category of the target detection object is determined according to the semantic data and the second detection data set.

[0091] Optionally, the processor 1001, in the execution of determining the behavior category of the target detection object according to the semantic data and the second detection data set, specifically performs the following operations: determining the action feature corresponding to the second detection data set according to the time sequence coding of each image data in the second detection data set; obtaining the semantic feature of the semantic data, the data dimensions of the semantic feature and the action feature being the same; determining the action behavior of the target detection object according to the action feature and the semantic feature, and identifying the action behavior to determine the behavior category of the target detection object.

[0092] Optionally, the processor 1001, in the execution of outputting the behavior category report of the target detection object if the behavior category is the abnormal category, specifically performs the following operations: matching the behavior category with each abnormal behavior in the preset abnormal behavior set; if the abnormal behavior set includes an abnormal behavior matching the behavior category, determining that the behavior category is the abnormal category, and generating and outputting the behavior category report of the target detection object based on the target location information and the behavior category.

[0093] Optionally, the processor 1001, in the execution of generating and outputting the behavior category report of the target detection object based on the target location information and the behavior category, specifically performs the following operations: determining a current time at which the first image data is obtained, generating the behavior category report of the target detection object based on the target location information, the behavior category and the current time, the behavior category report being used to describe the behavior category of the target detection object in the target location information; outputting the behavior category report on the display interface of the recording and broadcasting device.

[0094] In the embodiments of the present application, by obtaining the second image data and the first image data with higher resolution, and the audio data, after determining the target location information of the target detection object in the second image data, the region image data is intercepted in the first image data according to the target location information, the behavior of the target detection object is identified according to the semantic data corresponding to the region image data and the audio data, and the behavior category of the target detection object is determined, and the behavior category report is generated and output if the behavior category is the abnormal category, so that the target location information determined by the low-resolution second image data is used to identify the behavior category of the target detection object in the high-frequency first image data, and the accuracy and real-time performance of the behavior identification are ensured while reducing the data processing amount. Further, the region image data and the first historical image data are generated to generate the first detection data set according to the current time and the preset time length, so as to identify the behavior of the target detection object within a certain period of time, thereby improving the accuracy of the behavior category of the target detection object.

[0095] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiment methods can be included. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM), a random access memory (RAM), or the like.

[0096] The above only describes the preferred embodiments of the present application, and cannot limit the scope of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope of the present application.

Claims

1. A behavior recognition method, characterized in that, The method includes: Acquire first image data and second image data, as well as audio data, transmitted by the camera device, wherein the resolution of the first image data is higher than the resolution of the second image data; Identify the target detection object in the second image data, determine the target location information of the target detection object, and obtain the semantic data corresponding to the audio data, wherein the target detection object is any one of the detection objects in the second image data; Extract the region image data corresponding to the target location information from the first image data; Based on the semantic data and the region image data, the behavior category of the target detection object is determined; If the behavior category is an anomaly category, then output a behavior category report of the target detection object.

2. The method according to claim 1, characterized in that, The method is applied to a recording and broadcasting device, which is connected to a first computing power card and a camera device. The recording and broadcasting device includes a second computing power card, a display interface, and a data interface. The camera device is connected to the first computing power card and the second computing power card. The second computing power card is connected to the first computing power card through the data interface. The computing power of the first computing power card is higher than that of the second computing power card. The second computing card is used to identify the target detection object in the second image data, determine the target location information of the target detection object, and obtain the semantic data corresponding to the audio data, and transmit the target location information and the semantic data to the first computing card through the data interface; The first computing card is used to extract the region image data corresponding to the target location information from the first image data, and to determine the behavior category of the target detection object based on the semantic data and the region image data; If the behavior category is an abnormal category, then the display interface is controlled by the second computing card to output a behavior category report of the target detection object.

3. The method according to claim 1, characterized in that, The process of identifying the target detection object in the second image data, determining the target location information of the target detection object, and obtaining the semantic data corresponding to the audio data includes: The set of detected objects is obtained from the second image data based on a pre-trained object recognition model. The target detection object is located according to a pre-created positioning coordinate system to determine the target location information of the target detection object, wherein the target detection object is any detection object in the set of detection objects; The audio data is subjected to semantic recognition to obtain the semantic data corresponding to the audio data.

4. The method according to claim 1, characterized in that, The step of extracting the region image data corresponding to the target location information from the first image data includes: Determine the pixel ratio between the first image data and the second image data; The target position information is adjusted based on the pixel ratio to obtain the cropping position information; The first image data is extracted based on the extraction location information to obtain regional image data.

5. The method according to claim 1, characterized in that, Determining the behavior category of the target detection object based on the semantic data and the region image data includes: Acquire the first historical image data corresponding to the target detection object within a preset time period before the current time, wherein the current time is the time when the first image data was acquired; Based on the current time and the historical time of each of the first historical image data, sort each of the first historical image data and the regional image data to obtain a first detection data set arranged in chronological order; Based on the semantic data and the first detection data set, the behavior category of the target detection object is determined.

6. The method according to claim 5, characterized in that, The step of acquiring the first historical image data corresponding to the target detection object within a preset time period prior to the current moment includes: Acquire second historical image data within a preset time period prior to the current moment, wherein the resolution of the second historical image data is the same as the resolution of the first image data; Based on the target detection object, the second historical image data is cropped to obtain the first historical image data corresponding to the target detection object.

7. The method according to claim 5, characterized in that, Determining the behavior category of the target detection object based on the semantic data and the first detection data set includes: The data size of each image data in the first detection dataset is scaled to the target size to obtain the second detection dataset; The behavior category of the target detection object is determined based on the semantic data and the second detection data set.

8. The method according to claim 7, characterized in that, Determining the behavior category of the target detection object based on the semantic data and the second detection data set includes: Based on the temporal encoding of each image data in the second detection data set, determine the action features corresponding to the second detection data set; Obtain the semantic features of the semantic data, wherein the semantic features and the action features have the same data dimension; The action behavior of the target detection object is determined based on the action features and the semantic features, and the action behavior is identified to determine the behavior category of the target detection object.

9. The method according to claim 1, characterized in that, If the behavior category is an abnormal category, then a behavior category report of the target detection object is output, including: The behavior category is matched with each abnormal behavior in a preset set of abnormal behaviors; If the set of abnormal behaviors includes abnormal behaviors that match the behavior category, then the behavior category is determined to be an abnormal category, and a behavior category report of the target detection object is generated and output based on the target location information and the behavior category.

10. The method according to claim 9, characterized in that, The step of generating and outputting a behavior category report of the detected target object based on the target location information and the behavior category includes: Determine the current time when the first image data is acquired, and generate a behavior category report of the target detection object based on the target location information, the behavior category, and the current time. The behavior category report is used to describe the behavior category of the target detection object in the target location information. The behavior category report is output on the display interface of the recording device.

11. A behavior recognition system, characterized in that, The system includes a recording and broadcasting device, a first computing power card, and a camera device. The recording and broadcasting device includes a second computing power card, a display interface, and a data interface. The camera device is connected to the first computing power card and the second computing power card. The second computing power card is connected to the first computing power card through the data interface. The computing power of the first computing power card is higher than that of the second computing power card. The first computing card acquires the first image data transmitted by the camera device, and the second computing card acquires the second image data and audio data transmitted by the camera device, wherein the resolution of the first image data is higher than the resolution of the second image data; The second computing card identifies the target detection object in the second image data, determines the target location information of the target detection object, and obtains the semantic data corresponding to the audio data. The target location information and the semantic data are transmitted to the first computing card through the data interface. The target detection object is any one of the detection objects in the second image data. The first computing card extracts the region image data corresponding to the target location information from the first image data; The first computing card determines the behavior category of the target detection object based on the semantic data and the region image data, and transmits the behavior category to the second computing card through the data interface; If the behavior category is an abnormal category, the second computing card controls the display interface to output a behavior category report of the target detection object.

12. A computer storage medium storing a plurality of instructions adapted for loading by a processor and performing the steps of the method as claimed in any one of claims 1 to 10.

13. An electronic device, comprising: A processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to execute the steps of the method as claimed in any one of claims 1 to 10.

Images