Target recognition system and method based on machine vision

Abstract

The invention belongs to the technical field of image recognition. The invention provides a target recognition system and method based on machine vision. The method comprises the following steps: receiving feature description information of a specific target object input by a user, and performing semantic analysis on the feature description information by using a semantic analysis plug-in to obtain key feature data; retrieving a plurality of first target objects in the monitoring video based on the first static features, tracking each first target object for a preset duration, and extracting a second dynamic feature of each first target object from the tracking video; the similarity between each second dynamic feature and the first dynamic feature is calculated, the first target object corresponding to the second dynamic feature with the similarity higher than a similarity threshold is determined as a second target object, and the second target object is the specific target object. The specific target object can be recognized according to the feature description information of the voice and text types of the specific target object input by the user, and the image of the specific target object does not need to be obtained in advance.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a target recognition system and method based on machine vision. Background Art

[0002] Object recognition technology based on machine vision is an important branch of computer vision, which involves the task of identifying and locating specific targets from images or videos. For example, identifying people or objects with specific features in surveillance videos is a typical application scenario of object recognition technology.

[0003] When identifying and tracking people or objects with specific features based on surveillance videos, the images of people or objects with specific features are transmitted to the identification and tracking system, and then the identification and tracking system performs similarity analysis in the surveillance video based on the specific features extracted from the image. However, in most cases, it is very difficult to obtain images of people or objects, resulting in the inability to complete the identification and tracking of specific people or objects in surveillance videos, and only relying on inefficient identification by human eyes.

[0004] How to further improve target recognition based on machine vision is a technical problem that urgently needs to be solved. Summary of the invention

[0005] In response to the above technical problems, the present invention provides a machine vision-based target recognition method, system, electronic device, computer storage medium and computer program product to improve the efficiency and usability of machine vision-based target recognition.

[0006] The present invention discloses a target recognition method based on machine vision, which comprises the following steps: Receive feature description information of a specific target object input by a user, and use a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data; wherein the feature description information includes voice information and / or text information, and the key feature data includes a first static feature and a first dynamic feature; Retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data, tracking each of the first target objects for a preset time, and extracting a second dynamic feature of each of the first target objects from the tracking video; The similarity between each of the second dynamic features and the first dynamic feature is calculated, and the first target object corresponding to the second dynamic feature having a similarity higher than a similarity threshold is determined as the second target object, and the second target object is the specific target object.

[0007] Optionally, the receiving of feature description information of a specific target object input by a user, and performing semantic analysis on the feature description information using a semantic analysis plug-in to obtain key feature data includes: Receiving the feature description information of the specific target object input by a user, and performing format analysis on the feature description information; If the format analysis result only includes the image format, transmitting the feature description information to the target image recognition system; If the format analysis result includes voice format and / or text format, the feature description information is transmitted to the target semantic recognition system, and the target semantic recognition system uses a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain the key feature data.

[0008] Optionally, the using a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data includes: Using a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain a number of key feature words, analyzing the relevance between each of the key feature words and a specific target object, and determining each of the key feature words with a relevance higher than a relevance threshold as a target key feature word; The target key feature words are classified into static words and dynamic words, the first static feature is constructed according to the static words, the first dynamic feature is constructed according to the dynamic words, and the first static feature and the first dynamic feature are integrated to obtain the key feature data.

[0009] Optionally, retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data includes: Determine the object type and appearance features of the specific target object based on the first static features; A plurality of third target objects are retrieved in the surveillance video according to the object type, and each of the third target objects is screened secondary according to the appearance feature to obtain a plurality of the first target objects.

[0010] Optionally, the first dynamic feature includes first dynamic information of a specific target object and second dynamic information of an appendage associated with the specific target object; then each of the first target objects is tracked for a preset time period, and the second dynamic feature of each of the first target objects is extracted from the tracking video, including: Acquire the environment information of each of the first target objects, determine the preset time according to the environment information, and track each of the first target objects for the preset time; Based on the position change of the first target object in each video frame of the tracking video, the second dynamic feature of each first target object is calculated; wherein the second dynamic feature includes the third dynamic information of the first target object and the fourth dynamic information of the appendage associated with the first target object.

[0011] Optionally, the acquiring environment information of each of the first target objects and determining the preset duration according to the environment information includes: Acquire the environmental information of each of the first target objects, wherein the environmental information includes the density of the target objects and the moving speed of the target objects; The feature effective ratio is evaluated based on the target object density and the target object moving speed, and the preset time length is determined based on the feature effective ratio; wherein, the higher the feature effective ratio, the shorter the preset time length, and the lower the feature effective ratio, the longer the preset time length.

[0012] The present invention also discloses a target recognition system based on machine vision, the system comprising a processing device and a storage device, the computer code stored in the storage device is called and executed by the processing device to implement the following steps: Receive feature description information of a specific target object input by a user, and use a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data; wherein the feature description information includes voice information and / or text information, and the key feature data includes a first static feature and a first dynamic feature; Retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data, tracking each of the first target objects for a preset time, and extracting a second dynamic feature of each of the first target objects from the tracking video; The similarity between each of the second dynamic features and the first dynamic feature is calculated, and the first target object corresponding to the second dynamic feature having a similarity higher than a similarity threshold is determined as the second target object, and the second target object is the specific target object.

[0013] The present invention also discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement any of the above methods.

[0014] The present invention also discloses a computer storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement any of the above methods.

[0015] The present invention also discloses a computer program product, which includes computer codes. When the computer codes are executed by a processor of an electronic device, any of the above methods is implemented.

[0016] Compared with the existing target recognition methods recorded in the background technology, the present invention does not require users to obtain images of specific target objects. Instead, it can achieve fast and accurate recognition of specific target objects based on the feature description information of the voice and text types of the specific target objects input by the user, which can significantly reduce the difficulty of target recognition and improve adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flowchart of a method for target recognition based on machine vision disclosed in an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a machine vision-based target recognition system disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following is a description of the implementation of the present application by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0020] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0021] When existing surveillance videos are used to identify and track people or objects with specific features, the image of the person or object with specific features is transmitted to the identification and tracking system, and then the identification and tracking system performs similarity analysis in the surveillance video based on the specific features extracted from the image. However, in most cases, it is very difficult to obtain images of people or objects, resulting in the inability to complete the identification and tracking of specific people or objects in surveillance videos, and only relying on inefficient identification by the human eye. How to further improve target recognition based on machine vision is a technical problem that urgently needs to be solved.

[0022] In response to the above technical issues, such as Figure 1 As shown, an embodiment of the present invention discloses a target recognition method based on machine vision, and the method comprises the following steps: Receive feature description information of a specific target object input by a user, and use a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data; wherein the feature description information includes voice information and / or text information, and the key feature data includes a first static feature and a first dynamic feature; Retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data, tracking each of the first target objects for a preset time, and extracting a second dynamic feature of each of the first target objects from the tracking video; The semantic similarity between each of the second dynamic features and the first dynamic feature is calculated, and the first target object corresponding to the second dynamic feature having a semantic similarity higher than a similarity threshold is determined as the second target object, and the second target object is the specific target object.

[0023] In order to solve the above technical problems, the present invention provides feature description information of a specific target object that a user can input into the target recognition system. The feature description information can be voice information, text information, etc. The target recognition system then calls the semantic analysis plug-in to perform semantic analysis on the feature description information, and obtains key feature data from the analysis, including the first static feature and the first dynamic feature. Then, the target recognition system retrieves a number of first target objects in the surveillance video with the extracted first static feature, obtains a tracking video by tracking each first target object for a preset period of time, and extracts the second dynamic feature of each first target object from the tracking video. Finally, the semantic similarity between each second dynamic feature and the first dynamic feature is calculated, and the first target object corresponding to the second dynamic feature whose semantic similarity is higher than the similarity threshold is determined as the specific target object, thereby completing the recognition of the specific target object.

[0024] Compared with the existing target recognition methods recorded in the background technology, the present invention does not require users to obtain images of specific target objects. Instead, it can achieve fast and accurate recognition of specific target objects based on the feature description information of the voice and text types of the specific target objects input by the user, which can significantly reduce the difficulty of target recognition and improve adaptability.

[0025] Optionally, the receiving of feature description information of a specific target object input by a user, and performing semantic analysis on the feature description information using a semantic analysis plug-in to obtain key feature data includes: Receiving the feature description information of the specific target object input by a user, and performing format analysis on the feature description information; If the format analysis result only includes the image format, transmitting the feature description information to the target image recognition system; If the format analysis result includes voice format and / or text format, the feature description information is transmitted to the target semantic recognition system, and the target semantic recognition system uses a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain the key feature data.

[0026] In this embodiment, the present invention sets up two sets of target recognition systems, namely, a target image recognition system and a target semantic recognition system. Correspondingly, the user can input the image and / or feature description information of the specific target object. On this basis, the feature description information of the specific target object input by the user is first format analyzed. If the analysis shows that only the image format is included, the feature description information input by the user (i.e., the image of the specific target object) is transmitted to the target image recognition system, and the target image recognition system identifies the target object based on the conventional image similarity analysis method. If the analysis shows that the voice format and / or text format are included, the feature description information is transmitted to the target semantic recognition system, and the target semantic recognition system uses a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain the key feature data of the specific target object.

[0027] The system of the present invention can support input of relevant information of specific target features in various forms and has stronger adaptability.

[0028] Optionally, the using a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data includes: Using a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain a number of key feature words, analyzing the relevance between each of the key feature words and a specific target object, and determining each of the key feature words with a relevance higher than a relevance threshold as a target key feature word; The target key feature words are classified into static words and dynamic words, the first static feature is constructed according to the static words, the first dynamic feature is constructed according to the dynamic words, and the first static feature and the first dynamic feature are integrated to obtain the key feature data.

[0029] In this embodiment, the target semantic recognition system can call the semantic analysis plug-in to perform semantic analysis on the feature description information input by the user, and extract several key feature words therefrom. For example, if the feature description information is "I was running in the square just now and found a man wearing a black hat and carrying a white bag. The white bag felt very heavy. This person is very suspicious. Please find and track him." The extracted key feature words are: square, running, black hat, white bag, man, and heavy white bag.

[0030] Since the key feature words extracted above may contain words that are not related to the specific target object, it is necessary to remove the words that are not related to the specific target object. Specifically, the correlation between each key feature word and the specific target object is analyzed, and each key feature word with a correlation higher than the correlation threshold is determined as the target key feature word. For example, after analysis, it is found that the words "square", "black hat", "white bag", "man", and "white bag is heavy" have a high correlation with the specific target object, and they are determined as target key feature words.

[0031] Next, each target key feature vocabulary is classified into static vocabulary (square, black hat, white bag, man) and dynamic vocabulary (white bag is heavy), and the first static feature is constructed based on each static vocabulary, and the first dynamic feature is constructed based on each dynamic vocabulary, and the first static feature and the first dynamic feature are integrated to obtain key feature data. Among them, the first static feature can be each static vocabulary, or a sentence generated by the semantics of each static vocabulary, and correspondingly, the first dynamic feature can be each dynamic vocabulary, or a sentence generated by the semantics of each dynamic vocabulary.

[0032] It should be noted that the analysis of the correlation between each key feature word and a specific target object can be achieved through a large language model, such as a BERT large model. The BERT large model is based on the Transformer architecture and uses the self-attention mechanism to capture the relationship between words at different positions in the text, which enables the model to capture longer-distance dependencies. The present invention uses the BERT large model to analyze the correlation between key feature words and specific target objects. The BERT large model analyzes which key feature words in the feature description information input by the user are directly related to the specific target object, that is, determines which key feature words are directly related to the inputter of the feature description information.

[0033] Optionally, retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data includes: Determine the object type and appearance features of the specific target object based on the first static features; A plurality of third target objects are retrieved in the surveillance video according to the object type, and each of the third target objects is screened secondary according to the appearance feature to obtain a plurality of the first target objects.

[0034] In this embodiment, the first static feature includes the object type and appearance feature of the specific target object, the object type is, for example, a person, a man, a woman, a bag, a box, etc., and the appearance feature includes clothing, height, body shape, color, etc. First, a number of third target objects are retrieved from the surveillance video according to the object type, such as "man"; then, the third target objects are screened again according to the appearance feature, thereby obtaining a number of first target objects, such as "man wearing a black hat and holding a bag".

[0035] Optionally, the first dynamic feature includes first dynamic information of a specific target object and second dynamic information of an appendage associated with the specific target object; then each of the first target objects is tracked for a preset time period, and the second dynamic feature of each of the first target objects is extracted from the tracking video, including: Acquire the environment information of each of the first target objects, determine the preset time according to the environment information, and track each of the first target objects for the preset time; Based on the position change of the first target object in each video frame of the tracking video, the second dynamic feature of each first target object is calculated; wherein the second dynamic feature includes the third dynamic information of the first target object and the fourth dynamic information of the appendage associated with the first target object.

[0036] In this embodiment, the first target object is tracked for a preset time to obtain a corresponding tracking video, and the positions of the first target object and the accessories associated with the first target object are respectively identified in each video frame of the tracking video. By comparing the changes in the positions of the first target object and the accessories associated with the first target object in all video frames, the third dynamic information of the first target object and the fourth dynamic information of the accessories associated with the first target object are obtained. The third dynamic information and the fourth dynamic information constitute the second dynamic feature of the first target object (the second dynamic feature is, for example, the moving speed and swing speed / frequency of the first target object and its accessories, which can be a specific speed / frequency value or a set of original position information collected above). Correspondingly, the first dynamic feature obtained above also includes the first dynamic information of the specific target object and the second dynamic information of the accessories associated with the specific target object. Wherein, the accessories refer to handheld items of the first target object / specific target object, such as bags, boxes, etc.

[0037] Therefore, by calculating the semantic similarity between each second dynamic feature and the first dynamic feature, the semantic similarity between each first target object and the specific target object in terms of dynamic features can be analyzed. When the semantic similarity is higher than the similarity threshold (for example, the swing amplitude and frequency of the white bag meet the "white bag is heavy"), it means that the first target object is the specific target object to be identified and tracked. For example, when a man wearing a black hat is holding a heavy white bag, the swing amplitude and frequency of his arm will be at a very low level. Based on this dynamic feature, the specific target object can be further screened from each first target object.

[0038] Among them, the time span of a tracking video of the first target object obtained by tracking is dynamically set, and is specifically determined based on the environmental information of the first target object. In this way, the optimal time span can be determined based on the environmental information of the first target object, so that the second dynamic feature extracted subsequently can fully reflect the dynamic characteristics of the first target object, which is conducive to ensuring the confidence of the similarity obtained by analysis, thereby ensuring the recognition accuracy of the specific target object, and can also minimize the length of the tracking video and improve the response rate.

[0039] Optionally, the acquiring environment information of each of the first target objects and determining the preset duration according to the environment information includes: Acquire the environmental information of each of the first target objects, wherein the environmental information includes the density of the target objects and the moving speed of the target objects; A feature effective estimated value is obtained based on the target object density and the target object moving speed, and the preset time length is determined based on the feature effective estimated value; wherein, the higher the feature effective ratio, the shorter the preset time length, and the lower the feature effective ratio, the longer the preset time length.

[0040] In this embodiment, the environment information of the first target object can be extracted from the surveillance video, including the density of the target object and the moving speed of the target object. The target object may have the same attributes as the specific target object, for example, both are people. When the specific target object moves in the environment with the appendage in hand, the density of the target object and the moving speed of the target object in the environment will affect the swing amplitude and swing frequency of the hand of the specific target object. Specifically, when the density of the target object is higher and the moving speed of the target object is faster, the specific target object is not easy to swing the arm with a larger amplitude and a higher frequency, so that the swing amplitude and the swing frequency are at a lower level; and when the density of the target object is lower and the moving speed of the target object is slower, the specific target object is more likely to swing the arm with a larger amplitude and a higher frequency, so that the swing amplitude and the swing frequency are at a higher level (or normal level). At the same time, the density of the target object and the moving speed of the target object in the environment will also affect the "visibility" of the appendage of the specific target object in the surveillance video. Specifically, when the density of the target object is higher and the moving speed of the target object is faster, the probability of the appendage of the specific target object being blocked is higher, and its "visibility" in the surveillance video is worse.

[0041] The above two factors will cause the "visibility" of the specific target object and its accessories in the surveillance video to deteriorate, which will in turn cause the proximity between the extracted second dynamic feature and the normal level of swing data of the specific target object to deteriorate. It should be noted that the specific target object has both normal level swing and low level swing implemented to avoid other target objects in the entire surveillance video, but the higher the density of the target objects and the faster the target objects move, the lower the proportion of normal level swing in the entire surveillance video, so the average calculated second dynamic feature will have a greater deviation from the actual swing of the specific target object.

[0042] Moreover, the above two factors may also cause the extracted second dynamic features to contain dynamic features of fewer position points. For example, the actual swing amplitude of a specific target object is [-55°, +55°], but due to the high density of target objects in the environment and the fast movement speed of the target objects, the extracted second dynamic features have many position holes, such as the missing of multiple positions such as 40° and 20°. Obviously, this will cause the second dynamic feature to be significantly distorted, and the corresponding confidence of the calculated similarity will be insufficient, which is not conducive to the subsequent analysis to obtain a more accurate second target object, that is, a specific target object.

[0043] In view of the above technical problems, the present invention sets a feature effective estimation value according to the target object density in the environment where the first target object is located and the moving speed of the target object. The feature effective estimation value is used to characterize the proximity between the extracted second dynamic feature and the normal level swing data of the specific target object. The higher the feature effective estimation value, the higher the proximity, that is, the higher the credibility of the second dynamic feature. Based on the feature effective estimation value, the preset time length is determined according to the negative correlation relationship. Specifically, the higher the feature effective estimation value, the lower the target object density in the environment where the specific target object is located, the slower the moving speed of the target object, the lighter the occlusion and swing interference of the target object in the environment on the specific target object, and the higher the credibility of the extracted second dynamic feature (for example, the proportion of collected information at the same position is less, so that the position distribution range of each collected information will be larger, which is conducive to analyzing more accurate swing amplitude and frequency). At this time, the shorter the preset time length is set, the more accurate the second dynamic feature can be obtained without tracking for a longer time; on the contrary, the longer the preset time length is set, the more accurate the second dynamic feature can be obtained.

[0044] like Figure 2 As shown, an embodiment of the present invention further discloses a target recognition system based on machine vision, the system comprising a processing device and a storage device, the computer code stored in the storage device is called and executed by the processing device to implement the following steps: Receive feature description information of a specific target object input by a user, and use a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data; wherein the feature description information includes voice information and / or text information, and the key feature data includes a first static feature and a first dynamic feature; Retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data, tracking each of the first target objects for a preset time, and extracting a second dynamic feature of each of the first target objects from the tracking video; The similarity between each of the second dynamic features and the first dynamic feature is calculated, and the first target object corresponding to the second dynamic feature having a similarity higher than a similarity threshold is determined as the second target object, and the second target object is the specific target object.

[0045] An embodiment of the present invention further discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method described in the above embodiment.

[0046] An embodiment of the present invention further discloses a computer storage medium, wherein the computer storage medium stores a computer program, and the computer program is executed by a processor to implement the method described in the above embodiment.

[0047] An embodiment of the present invention further discloses a computer program product, which includes computer code. When the computer code is executed by a processor of an electronic device, the method described in the above embodiment is implemented.

[0048] The computer-readable storage medium described above may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0049] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0050] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0051] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A target recognition method based on machine vision, characterized in that: The method comprises the following steps: Receive feature description information of a specific target object input by a user, and use a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data; wherein the feature description information includes voice information and / or text information, and the key feature data includes a first static feature and a first dynamic feature; Retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data, tracking each of the first target objects for a preset time, and extracting a second dynamic feature of each of the first target objects from the tracking video; The similarity between each of the second dynamic features and the first dynamic feature is calculated, and the first target object corresponding to the second dynamic feature having a similarity higher than a similarity threshold is determined as the second target object, and the second target object is the specific target object.

2. The target recognition method based on machine vision according to claim 1, characterized in that: Receive feature description information of a specific target object input by a user, use a semantic analysis plug-in to perform semantic analysis on the feature description information, and obtain key feature data, including: Receiving the feature description information of the specific target object input by a user, and performing format analysis on the feature description information; If the format analysis result only includes the image format, transmitting the feature description information to the target image recognition system; If the format analysis result includes voice format and / or text format, the feature description information is transmitted to the target semantic recognition system, and the target semantic recognition system uses a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain the key feature data.

3. The target recognition method based on machine vision according to claim 2, characterized in that: Use the semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data, including: Using a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain a number of key feature words, analyzing the relevance between each of the key feature words and a specific target object, and determining each of the key feature words with a relevance higher than a relevance threshold as a target key feature word; The target key feature words are classified into static words and dynamic words, the first static feature is constructed according to the static words, the first dynamic feature is constructed according to the dynamic words, and the first static feature and the first dynamic feature are integrated to obtain the key feature data.

4. The target recognition method based on machine vision according to claim 3, characterized in that: Retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data includes: Determine the object type and appearance features of the specific target object based on the first static features; A plurality of third target objects are retrieved in the surveillance video according to the object type, and each of the third target objects is screened secondary according to the appearance feature to obtain a plurality of the first target objects.

5. The target recognition method based on machine vision according to claim 4, characterized in that: The first dynamic feature includes first dynamic information of a specific target object and second dynamic information of an appendage associated with the specific target object; each of the first target objects is tracked for a preset time, and the second dynamic feature of each of the first target objects is extracted from the tracking video, including: Acquire the environment information of each of the first target objects, determine the preset time period according to the environment information, and track each of the first target objects for the preset time period; Based on the position change of the first target object in each video frame of the tracking video, the second dynamic feature of each first target object is calculated; wherein the second dynamic feature includes the third dynamic information of the first target object and the fourth dynamic information of the appendage associated with the first target object.

6. The target recognition method based on machine vision according to claim 5, characterized in that: Acquiring environmental information of each of the first target objects, and determining the preset duration according to the environmental information, includes: Acquire the environmental information of each of the first target objects, wherein the environmental information includes the density of the target objects and the moving speed of the target objects; The feature effective ratio is evaluated based on the target object density and the target object moving speed, and the preset time length is determined based on the feature effective ratio; wherein, the higher the feature effective ratio, the shorter the preset time length, and the lower the feature effective ratio, the longer the preset time length.

7. A target recognition system based on machine vision, the system comprising a processing device and a storage device, characterized in that: The computer code stored in the storage device is called and executed by the processing device to implement the following steps: Receive feature description information of a specific target object input by a user, and use a semantic analysis plug-in to perform semantic analysis on the feature description information to obtain key feature data; wherein the feature description information includes voice information and / or text information, and the key feature data includes a first static feature and a first dynamic feature; Retrieving a plurality of first target objects in the surveillance video based on the first static feature in the key feature data, tracking each of the first target objects for a preset time, and extracting a second dynamic feature of each of the first target objects from the tracking video; The similarity between each of the second dynamic features and the first dynamic feature is calculated, and the first target object corresponding to the second dynamic feature having a similarity higher than a similarity threshold is determined as the second target object, and the second target object is the specific target object.

8. An electronic device comprising: At least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 6.

9. A computer storage medium storing a computer program, characterized in that: The computer program is executed by a processor to implement the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that: The computer program product includes computer codes, and when the computer codes are executed by a processor of an electronic device, the method according to any one of claims 1 to 6 is implemented.

Images