Human false detection filtering method and system based on multi-dimensional comparison and medium

Abstract

The application provides a human body false detection filtering method and system based on multi-dimensional comparison, a medium, and the like. First, the target confidence result is compared with a preset false detection library execution threshold value. When the target confidence result is greater than or equal to the preset false detection library execution threshold value, it is directly determined that an alarm is generated, and a subsequent comparison process is not needed, thereby improving detection efficiency. When the target confidence result is less than the preset false detection library execution threshold value, a false detection filtering algorithm is used, and a false detection image library constructed in advance is used to perform multi-dimensional determination of human body feature factor double-threshold screening, false detection library feature similarity comparison, and size similarity comparison. The method does not need to rely on massive labeled data to expand a training set, thereby reducing the cost of data collection, labeling, and model training, reducing the false detection rate and the missed detection rate, and improving the reliability and scene adaptability of an alarm determination result.

Description

Technical Field

[0001] This application relates to the field of target detection technology, and in particular to a method, system and medium for filtering false detections of human bodies based on multi-dimensional comparison. Background Technology

[0002] In industrial safety production, preventing unauthorized entry into hazardous equipment operating areas is a crucial aspect of ensuring personal safety and maintaining production continuity. To achieve this goal, the industry widely employs artificial intelligence-based target detection algorithms to analyze images captured by surveillance cameras in real time, determine the presence of personnel within pre-defined hazardous areas, trigger corresponding alarm signals, and control the start / stop of hazardous equipment within the area accordingly—when a person is detected, the hazardous equipment is stopped to prevent injury; when no one is detected, the hazardous equipment is allowed to start or continue operating normally. This type of personnel detection technology is of great significance in ensuring personal safety and production continuity.

[0003] However, commonly used personnel detection algorithms in existing technologies are limited by the characteristics of the algorithm models themselves and the coverage of training data, and generally have a certain risk of false detection, that is, they are prone to misidentifying non-human targets in the monitoring screen as human beings. Such false detection problems not only cause frequent false alarms, leading to abnormal start-ups and shutdowns of production equipment, but may also cause a decrease in production line efficiency and even affect the overall safety management level of the factory.

[0004] Existing solutions for reducing false positives typically rely on increasing the amount of training data or using region masking, mainly including the following:

[0005] (1) Filtering scheme based on a single classification network: After the preceding target detection algorithm detects a suspected human target, the suspected target is input into a trained classification network, which classifies it into two categories: "human" or "non-human," thereby achieving false detection filtering. However, this scheme usually relies on a deep learning model based on a convolutional neural network and requires a large amount of data for training. If the amount of data is insufficient, the model's generalization ability will be insufficient, resulting in some classification errors in new scenarios, which in turn leads to certain false detections and false negatives. Although retraining the classification network can alleviate this problem to some extent, retraining requires recollecting scene data, labeling the data, and spending a lot of time retraining, resulting in high training costs and making it impossible to achieve rapid and stable detection of people in dangerous areas.

[0006] (2) False Detection Filtering Scheme Based on Region Masking: After the target detection algorithm detects a target, it outputs six values ​​for a single target: category, x-coordinate, y-coordinate, width, height, and confidence level. If the target is a false detection, a masking box with the same width and height can be drawn at the location of the detected target to completely cover the target bounding box. An IOU (Intersection Over Union) threshold is preset, which is the degree of overlap between two rectangular boxes, with a value range of 0-1.0. When the IOU between the subsequently detected target and the masking box is greater than the preset threshold, it is judged as a false detection, and no alarm is output. Although this method can completely mask false detection targets of similar size at specific locations in the monitoring area, it will also mask normal human targets of similar size at the same location, which poses a certain risk of missed detection. It lacks flexibility and cannot adapt to complex and ever-changing actual scenarios.

[0007] (3) Confidence-based filtering scheme: Each target detected by the target detection model usually corresponds to a confidence score, which represents the probability that the model determines the predicted target to be a "person". The value range is usually 0-1.0. This scheme filters out false detections by setting a confidence threshold (such as 0.6) and only outputting targets with confidence scores higher than the threshold as the final detection result. However, in actual industrial scenarios, some detected real human targets may have low confidence scores due to occlusion or other reasons, falling below the preset threshold and thus being filtered out, leading to missed detections. At the same time, some non-human targets may have confidence scores higher than the preset threshold due to their similar appearance to humans, and cannot be effectively filtered out, still resulting in false detections. Therefore, a filtering scheme based solely on confidence cannot fundamentally eliminate the problem of false detections.

[0008] Therefore, how to further reduce the false detection rate, avoid abnormal equipment start-up and shutdown, and improve production efficiency while maintaining an extremely low false detection rate and ensuring personnel safety has become a key technical problem that is difficult to overcome and urgently needs to be solved in the existing personnel detection technology for industrial safety production. Summary of the Invention

[0009] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a method, system and medium for filtering false detections of human bodies based on multi-dimensional comparison, so as to solve the technical problems of false detection risk and false negative detection risk in existing human target detection algorithms.

[0010] To achieve the above and other related objectives, a first aspect of this application provides a method for filtering false human detections based on multi-dimensional comparison, comprising: acquiring an image of a preset danger zone to be identified; performing a target identification operation based on the image to be identified to generate a target determination result; wherein the target determination result includes a target confidence result; and generating an alarm determination result based on the target determination result, and on a false detection filtering algorithm and a preset false detection library execution threshold; the method includes: if the target confidence result is greater than or equal to the preset false detection library execution threshold, then determining the alarm determination result as an output alarm; if the target confidence result is less than the preset false detection library execution threshold, then performing a multi-dimensional similarity comparison operation based on the false detection filtering algorithm and using a pre-constructed false detection image library to generate an alarm determination result.

[0011] In some embodiments of the first aspect of this application, the method of generating an alarm judgment result by performing a multi-dimensional similarity comparison operation based on a false detection filtering algorithm and utilizing a pre-built false detection image library includes: performing a cropping operation on the image to be identified to generate a target image to be judged; performing an inference operation on the target image to be judged based on a human feature extraction network to generate human feature factors to be judged; and performing a multi-dimensional similarity comparison operation based on the human feature factors to be judged, a preset high threshold for human feature factors, a preset low threshold for human feature factors, and the false detection image library to generate an alarm judgment result.

[0012] In some embodiments of the first aspect of this application, the method of generating an alarm determination result by performing a multi-dimensional similarity comparison operation based on the human feature factor to be determined, a preset high threshold for human feature factors, a preset low threshold for human feature factors, and the false detection image library includes: if the human feature factor to be determined is greater than or equal to the preset high threshold for human feature factors, then the alarm determination result is determined to be an alarm output; if the human feature factor to be determined is less than the preset high threshold for human feature factors, then it is determined whether the human feature factor to be determined is less than or equal to the preset low threshold for human feature factors; if the human feature factor to be determined is less than or equal to the preset low threshold for human feature factors, then the alarm determination result is determined to be no alarm; if the human feature factor to be determined is greater than the preset low threshold for human feature factors and less than the preset high threshold for human feature factors, then a multi-dimensional similarity comparison operation is performed based on the false detection image library to generate an alarm determination result.

[0013] In some embodiments of the first aspect of this application, the false detection image library includes multiple non-human false detection sample images; wherein, the method of performing a multi-dimensional similarity comparison operation based on the false detection image library to generate an alarm judgment result includes: performing a human feature similarity comparison operation based on the human feature factor to be judged to calculate a human feature false detection similarity score; performing a false detection feature similarity comparison operation and a false detection size similarity comparison operation based on the target image to be judged and each non-human false detection sample image in the false detection image library to calculate a false detection feature similarity score and a false detection size similarity score; calculating a final target similarity score based on the human feature false detection similarity score, the false detection feature similarity score and the false detection size similarity score; and generating an alarm judgment result based on the final target similarity score and a preset similarity score threshold.

[0014] In some embodiments of the first aspect of this application, the method of performing a false detection feature similarity comparison operation and a false detection size similarity comparison operation on the target image to be determined and each non-human false detection sample image in the false detection image library to calculate the false detection feature similarity score and the false detection size similarity score includes: performing a first feature extraction operation on the target image to be determined based on the false detection library model to generate a target feature vector and a target size vector; performing a second feature extraction operation on each of the non-human false detection sample images based on the false detection library model to generate multiple false detection feature vectors and multiple false detection size vectors; wherein each non-human false detection sample image corresponds to one false detection feature vector and one false detection size vector; performing a false detection feature similarity comparison operation on the target feature vector and each of the false detection feature vectors to lock the false detection matching image and the optimal matching size vector, and calculating the false detection feature similarity score; and performing a false detection size similarity comparison operation on the target size vector and the optimal matching size vector to calculate the false detection size similarity score.

[0015] In some embodiments of the first aspect of this application, the method of performing a false detection feature similarity comparison operation based on the target feature vector and each of the false detection feature vectors to lock the false detection matching image and the optimal matching size vector, and calculating the false detection feature similarity score includes: performing a false detection feature similarity comparison operation based on the target feature vector and each of the false detection feature vectors to calculate the Euclidean distance between the target feature vector and each of the false detection feature vectors, and taking the Euclidean distance with the smallest value as the optimal matching Euclidean distance; locking the non-human false detection sample image corresponding to the optimal matching Euclidean distance in the false detection image library as the false detection matching image, and taking the false detection size vector of the non-human false detection sample image corresponding to the optimal matching Euclidean distance as the optimal matching size vector; and calculating the false detection feature similarity score based on the optimal matching Euclidean distance.

[0016] In some embodiments of the first aspect of this application, the target size vector includes a target width vector and a target height vector; the optimal matching size vector includes a matching width vector and a matching height vector; wherein, the method of performing a false detection size similarity comparison operation based on the target size vector and the optimal matching size vector to calculate a false detection size similarity score includes: calculating a false detection width difference metric based on the target width vector and the matching width vector; calculating a false detection height difference metric based on the target height vector and the matching height vector; calculating a false detection width similarity score based on the false detection width difference metric; calculating a false detection height similarity score based on the false detection height difference metric; and calculating a false detection size similarity score based on the false detection width similarity score and the false detection height similarity score.

[0017] In some embodiments of the first aspect of this application, the final target similarity score is calculated based on the false detection similarity score of the human body features, the false detection feature similarity score, and the false detection size similarity score, including the following methods:

[0018] ;

[0019] in, This represents the final target similarity score; Indicates the weight of the similarity score for false positive databases; This represents the similarity score of false positive features; Indicates the weight of human characteristic factors; This represents the similarity score for false positives of human features. Indicates the weight of the size similarity score; This represents the false positive size similarity score.

[0020] To achieve the above and other related objectives, a second aspect of this application provides a human false detection filtering system based on multi-dimensional comparison, comprising: an image acquisition module for acquiring an image of a preset danger area to be identified; a target identification module for performing a target identification operation based on the image to be identified to generate a target determination result; wherein the target determination result includes a target confidence result; and a false detection filtering module for generating an alarm determination result based on the target determination result, a false detection filtering algorithm, and a preset false detection library execution threshold; wherein the method includes: if the target confidence result is greater than or equal to the preset false detection library execution threshold, then determining the alarm determination result as an output alarm; if the target confidence result is less than the preset false detection library execution threshold, then performing a multi-dimensional similarity comparison operation based on the false detection filtering algorithm and using a pre-constructed false detection image library to generate an alarm determination result.

[0021] To achieve the above and other related objectives, a third aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the human body false detection filtering method based on multi-dimensional comparison as described above.

[0022] As described above, the human false detection filtering method, system, and medium based on multi-dimensional comparison of this application have the following beneficial effects:

[0023] (1) Layered judgment design to improve detection efficiency and judgment accuracy: First, the target confidence result is compared with the preset false detection library execution threshold. When the target confidence result is greater than or equal to the preset false detection library execution threshold, an alarm is directly determined without entering the subsequent comparison process, thus improving detection efficiency. When the target confidence result is less than the preset false detection library execution threshold, based on the false detection filtering algorithm and using the pre-built false detection image library, a multi-dimensional judgment is performed, including human feature factor dual threshold screening, false detection library feature similarity comparison, and size similarity comparison. This not only avoids the missed detection of real human targets under occlusion and other conditions, but also effectively filters out non-human false detection targets with similar appearance, reducing the false detection rate and the missed detection rate.

[0024] (2) Reduce false detection rate and improve judgment reliability: By using the human feature false detection similarity score, the possibility of false detection is judged from the dimension of human feature similarity. By using the false detection feature similarity score and the false detection size similarity score, the images are compared with each non-human false detection sample image in the false detection image library from the dimensions of false detection image feature similarity and size similarity. Only non-human targets that are highly matched with the features and size of the false detection library samples are filtered and fused to obtain the final target similarity score. Real human targets are not blocked, thus realizing the distinction between non-human false detection targets and real human targets, reducing the false detection rate, ensuring the accuracy of false detection filtering, and improving the reliability and scene adaptability of alarm judgment results.

[0025] (3) Human feature dual threshold screening to reduce invalid comparisons: When the target confidence result is less than the preset false detection library execution threshold, the human feature factor to be judged is first compared with the preset human feature factor high threshold and preset human feature factor low threshold. This can quickly distinguish highly credible human, highly credible non-human and fuzzy targets. Only fuzzy targets whose human feature factor to be judged is greater than the preset human feature factor low threshold and less than the preset human feature factor high threshold are subjected to subsequent false detection library feature comparison. This can quickly filter a large number of clear targets, reduce invalid feature comparison and calculation operations, further improve the algorithm running efficiency, and avoid misjudgment of clear targets, reducing the risk of missed detection and false detection.

[0026] (4) Reduce training costs: False detection filtering can be achieved by comparing the preset threshold with the false detection image library. There is no need to rely on massive labeled data to expand the training set, and there is no need to frequently retrain the model. This reduces the cost of data collection, labeling and model training. The real-time update of the false detection image library can adapt to new scenarios and new interference targets, and achieve rapid and stable detection of people in dangerous areas. Attached Figure Description

[0027] Figure 1 The diagram shown is a flowchart of a human false detection filtering method based on multi-dimensional comparison in one embodiment of this application.

[0028] Figure 2 The diagram shown is a flowchart of a false detection filtering algorithm in one embodiment of this application.

[0029] Figure 3 The diagram shows a flowchart of performing a multi-dimensional similarity comparison operation in one embodiment of this application.

[0030] Figure 4 This is a schematic diagram illustrating another process for performing multi-dimensional similarity comparison operations in one embodiment of this application.

[0031] Figure 5 The diagram shows a flowchart of the similarity comparison of false detection samples in one embodiment of this application.

[0032] Figure 6 The diagram shows the structure of the original ResNet18 convolutional neural network.

[0033] Figure 7 The diagram shown is a schematic block diagram of a human false detection filtering system based on multi-dimensional comparison in one embodiment of this application. Detailed Implementation

[0034] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0035] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" do not necessarily imply that they are different.

[0036] It should be noted that, in the embodiments of this application, the words "exemplary" or "for example" refer to examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0037] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0038] Before providing a further detailed description of the present invention, the nouns and terms used in the embodiments of the present invention are explained, and the nouns and terms used in the embodiments of the present invention are subject to the following interpretations:

[0039] <1> ResNet18: Residual Network with 18 layers, is a deep convolutional neural network with an 18-layer network structure (17 convolutional layers + 1 fully connected layer).

[0040] To facilitate understanding of the embodiments of this application, firstly, in conjunction with Figure 1 Detailed explanation. Figure 1 This document illustrates a flowchart of a human false detection filtering method based on multi-dimensional comparison, as described in an embodiment of the present invention. The method mainly includes the following steps:

[0041] S101: Acquire images of the preset hazardous areas to be identified.

[0042] In this embodiment, during industrial production, hazardous industrial equipment operates by using monitoring cameras deployed in designated hazardous areas to capture real-time images of these areas, thereby obtaining images of targets to be identified. These images are real-time monitoring images of the hazardous areas within the industrial site, used for subsequent target detection and human identification.

[0043] S102: Based on the image to be identified, perform a target recognition operation to generate a target determination result; wherein the target determination result includes a target confidence result.

[0044] In this embodiment, the image to be identified is input into a preceding detection model to output a target determination result, which includes:

[0045] (1) Category result: indicates the classification judgment of the target in the image to be identified by the preceding detection model, specifically "human" or "non-human".

[0046] (2) Target position results: including vertical coordinate, horizontal coordinate, width and height results, used to represent the position and size of the detected target in the image to be identified.

[0047] (3) Target confidence result: indicates the probability that the preceding detection model determines the target to be a "person", with a value range of 0 to 1.0.

[0048] In this embodiment, the preliminary detection model is a deep learning-based object detection model, typically constructed using a convolutional neural network structure. After training on a large number of samples, this model is used for preliminary detection and classification of targets in the image to be identified.

[0049] S103: Based on the target determination result, and using a false detection filtering algorithm and a preset false detection library execution threshold, an alarm determination result is generated; the method includes:

[0050] (1) If the target confidence result is greater than or equal to the preset false detection library execution threshold, then the alarm determination result is determined to be an output alarm.

[0051] (2) If the target confidence result is less than the preset false detection library execution threshold, then based on the false detection filtering algorithm and using the pre-built false detection image library, a multi-dimensional similarity comparison operation is performed to generate an alarm judgment result.

[0052] In this embodiment, it is determined whether the target confidence result is greater than or equal to the preset false detection library execution threshold. If the target confidence result is greater than or equal to the preset false detection library execution threshold, the current detection target is directly determined as a valid human target, and a corresponding alarm judgment result is generated. This alarm judgment result is an output alarm to control the dangerous industrial equipment to stop operating, thereby realizing the safety protection of on-site personnel.

[0053] In this embodiment, as Figure 2 The diagram illustrates the flowchart of the false detection filtering algorithm in this embodiment of the invention. If the target confidence result is less than the preset false detection library execution threshold, a multi-dimensional similarity comparison operation is performed based on the false detection filtering algorithm and using a pre-built false detection image library to generate an alarm judgment result. The method includes:

[0054] S201: Perform a cropping operation on the image to be identified to generate a target image to be judged.

[0055] In this embodiment, for images to be identified whose target confidence result is less than the preset false detection threshold, a cropping operation is performed to generate a target image t to be judged.

[0056] S202: Based on the human feature extraction network, perform inference operations on the target image to be judged to generate human feature factors to be judged.

[0057] In this embodiment, the human feature extraction network (F_Cls) is a pre-trained convolutional neural network model used to extract human-related features from the target image to be judged and output a confidence feature value representing the target as a human body, i.e., the human feature factor Cls_t to be judged. This human feature extraction network (F_Cls) is trained on a large image dataset containing human and non-human samples, and has the ability to extract and classify features of human regions. It can effectively distinguish human regions from background and interference regions, providing human feature basis for subsequent multi-dimensional joint judgment.

[0058] S203: Based on the human feature factors to be determined, and based on the preset high threshold of human feature factors, the preset low threshold of human feature factors, and the false detection image library, perform a multi-dimensional similarity comparison operation to generate an alarm determination result.

[0059] In this embodiment, as Figure 3 The diagram illustrates a flowchart of a multi-dimensional similarity comparison operation performed in an embodiment of the present invention. Based on the human feature factors to be determined, and using preset high thresholds and low thresholds for human feature factors, along with the false detection image library, a multi-dimensional similarity comparison operation is performed to generate an alarm determination result.

[0060] S301: If the human characteristic factor to be determined is greater than or equal to the preset high threshold of human characteristic factor, then the alarm determination result is determined to be an output alarm.

[0061] S302: If the human characteristic factor to be determined is less than the preset high threshold of human characteristic factor, then determine whether the human characteristic factor to be determined is less than or equal to the preset low threshold of human characteristic factor.

[0062] S303: If the human characteristic factor to be determined is less than or equal to the preset low threshold of human characteristic factor, then the alarm determination result is determined to be no alarm.

[0063] S304: if the human feature factor to be determined is greater than the low threshold of the preset human feature factor and less than the high threshold of the preset human feature factor, performing a multi-dimensional similarity comparison operation based on the false detection image library to generate an alarm determination result.

[0064] In this embodiment, the preset high threshold of human feature factor T_high is used to determine highly credible human targets, and the preset low threshold of human feature factor T_low is used to determine obvious non-human targets. The human feature factor to be determined Cls_t is compared with the preset high threshold of human feature factor T_high (e.g., 10.0) and the preset low threshold of human feature factor T_low (e.g., 2.0). When the human feature factor to be determined is greater than or equal to the preset high threshold of human feature factor (i.e., Cls_t≥T_high), it indicates that the current detection target has significant human features, can be directly determined as a human body, and a corresponding alarm determination result is generated. The alarm determination result is to output an alarm, trigger a shutdown, and control hazardous industrial equipment to stop operating, thereby realizing safety protection for on-site personnel.

[0065] In this embodiment, when the human feature factor to be determined Cls_t is less than or equal to the preset low threshold of human feature factor T_low (i.e., Cls_t≤T_low), it indicates that the current detection target does not have human features and belongs to a false detection, can be directly determined as a non-human body, and a corresponding alarm determination result is generated. The alarm determination result is that no alarm is given, and the hazardous industrial equipment continues to operate.

[0066] In this embodiment, when the human feature factor to be determined Cls_t is greater than the preset low threshold of human feature factor T_low and less than the preset high threshold of human feature factor T_high (i.e., T_low<Cls_t<T_high), it indicates that the human features of the current detection target are unclear, and it cannot be directly determined as a credible human body nor directly determined as an obvious non-human body. It belongs to an intermediate area target with fuzzy human features that is prone to false detection, then the subsequent multi-dimensional joint determination process of feature similarity comparison and size similarity based on the false detection image library is entered to further distinguish human targets from false detection targets.

[0067] It is worth noting that through the dual-threshold screening of human features, only the fuzzy targets whose human feature factor to be determined is greater than the low threshold of the preset human feature factor and less than the high threshold of the preset human feature factor are subjected to subsequent feature comparison with the false detection library, so that a large number of clear targets can be quickly filtered out, invalid feature comparison and calculation operations are reduced, the operation efficiency of the algorithm is further improved, misjudgment of clear targets is avoided, and the risk of missing detection and false detection is reduced.

[0068] In this embodiment, for example Figure 4The diagram illustrates another flowchart illustrating the multi-dimensional similarity comparison operation in an embodiment of the present invention. The multi-dimensional similarity comparison operation includes human feature similarity comparison, false detection feature similarity comparison, and false detection size similarity comparison. The false detection image library includes multiple non-human false detection sample images; wherein, based on the false detection image library, the multi-dimensional similarity comparison operation is performed to generate an alarm judgment result, including:

[0069] S401: Based on the human feature factors to be determined, perform a human feature similarity comparison operation to calculate the human feature false detection similarity score.

[0070] In this embodiment, the method for calculating the false detection similarity score of human features includes:

[0071] Formula (1)

[0072] in, Cls_t represents the similarity score of false positives for human features; Cls_t represents the human feature factor to be determined.

[0073] In this embodiment, the human feature false detection similarity score Used to characterize the confidence level that the current detection target is a non-human false positive. Human feature false positive similarity score. The higher the score, the weaker the human characteristics of the target being detected, and the higher the probability of false detection. Human Feature False Detection Similarity Score The lower the value, the stronger the human characteristics of the target being detected, and the lower the possibility of false detection.

[0074] S402: Based on the target image to be determined and each non-human false detection sample image in the false detection image library, perform false detection feature similarity comparison operation and false detection size similarity comparison operation to calculate the false detection feature similarity score and the false detection size similarity score.

[0075] In this embodiment, the false detection image library includes multiple non-human false detection sample images. The non-human false detection sample images are non-human target images that were mistakenly identified as human by the preceding detection model. The purpose of reading each non-human false detection sample image is to extract the features of the false detection target, which will then be used to compare with the real-time detected target in order to identify similar false detections.

[0076] In this embodiment, as Figure 5 The diagram illustrates the flowchart of false detection sample similarity comparison in an embodiment of the present invention. Based on the target image to be determined and various non-human false detection sample images in the false detection image library, false detection feature similarity comparison and false detection size similarity comparison operations are performed to calculate the false detection feature similarity score and the false detection size similarity score.

[0077] S4021: Based on the false detection library model, perform a first feature extraction operation on the target image to be determined to generate a target feature vector and a target size vector.

[0078] In this embodiment, as Figure 6 The diagram shows the structure of the original ResNet18 convolutional neural network. The false positive database model (F_Sim) is a feature extraction model based on the ResNet18 convolutional neural network, used to extract high-dimensional feature vectors from images. The original ResNet18 convolutional neural network contains 17 convolutional layers and 1 fully connected layer for classification (i.e., ...). Figure 6 (FC in the original text). In this invention, the false detection library model (F_Sim) removes this fully connected layer, retaining only a 17-convolutional layer feature extraction backbone network, and discarding the classification branch to focus on the general feature representation of the image. The false detection library model (F_Sim) takes an integer image data of size 1*3*224*224 as input and outputs a 1*512 floating-point data, representing the feature vector of that image.

[0079] In this embodiment, the false detection library model (F_Sim) is used to extract features from the target image to be judged, output the target feature vector corresponding to the target image to be judged, and simultaneously record the size information of the target image to be judged to generate the target size vector.

[0080] S4022: Based on the false detection library model, perform a second feature extraction operation on each of the non-human false detection sample images to generate multiple false detection feature vectors and multiple false detection size vectors; wherein, each non-human false detection sample image corresponds to one false detection feature vector and one false detection size vector.

[0081] In this embodiment, a false detection library model (F_Sim) is used to extract features from each non-human false detection sample image in the false detection image library. For each non-human false detection sample image, a corresponding false detection feature vector is output, and the size information of each non-human false detection sample image is recorded simultaneously to generate a corresponding false detection size vector. Both the target feature vector and the false detection feature vector are 512-dimensional feature vectors.

[0082] S4023: Based on the target feature vector and each of the false detection feature vectors, perform a false detection feature similarity comparison operation to lock the false detection matching image and the optimal matching size vector, and calculate the false detection feature similarity score. The method includes:

[0083] (1) Based on the target feature vector and each of the false detection feature vectors, perform a false detection feature similarity comparison operation to calculate the Euclidean distance between the target feature vector and each of the false detection feature vectors, and take the Euclidean distance with the smallest value as the optimal matching Euclidean distance.

[0084] In this embodiment, the target feature vector includes multiple first feature vectors, and the false detection feature vector includes multiple second feature vectors. The method for calculating the Euclidean distance between the target feature vector and the false detection feature vector includes:

[0085] Formula (II)

[0086] Where d represents the Euclidean distance between the target feature vector and the false detection feature vector; This represents the nth first feature vector in the target feature vector; This represents the nth second feature vector in the false detection feature vector.

[0087] In this embodiment, the Euclidean distances between the target feature vector and each false detection feature vector are compared. The larger the Euclidean distance, the lower the similarity. The Euclidean distance with the smallest value is taken as the optimal matching Euclidean distance, and the non-human false detection sample image corresponding to the smallest Euclidean distance is locked.

[0088] (2) In the false detection image library, the non-human false detection sample image corresponding to the optimal matching Euclidean distance is locked as the false detection matching image, and the false detection size vector of the non-human false detection sample image corresponding to the optimal matching Euclidean distance is used as the optimal matching size vector.

[0089] In this embodiment, the non-human false detection sample image corresponding to the optimal matching Euclidean distance is locked as the false detection matching image, and the false detection size vector of the non-human false detection sample image is used as the optimal matching size vector of the false detection matching image.

[0090] (3) Calculate the false detection feature similarity score based on the optimal matching Euclidean distance. The method includes:

[0091] Formula (3)

[0092] in, This represents the similarity score of false positive features; This indicates the hyperparameter for the similarity between the false detection databases; represents the optimal matching Euclidean distance; exp represents the natural exponential function.

[0093] In this embodiment, the false detection feature similarity score The value ranges from 0 to 1.0. The hyperparameter for false positive database similarity. It is 0.5.

[0094] S4024: Based on the target size vector and the optimal matching size vector, perform a false detection size similarity comparison operation to calculate the false detection size similarity score.

[0095] In this embodiment, the target size vector includes a target width vector and a target height vector; the optimal matching size vector includes a matching width vector and a matching height vector; wherein, the method of performing a false detection size similarity comparison operation based on the target size vector and the optimal matching size vector to calculate the false detection size similarity score includes:

[0096] (1) The false detection width difference metric is calculated based on the target width vector and the matching width vector.

[0097] In this embodiment, the width ratio between the target image to be determined and the falsely detected matching image is calculated based on the target width vector of the target image to be determined and the matching width vector of the falsely detected matching image. The method includes:

[0098] Formula (IV)

[0099] in, This represents the width ratio between the target image to be judged and the falsely detected matching image; Represents the target width vector; This indicates the matching width vector.

[0100] In this embodiment, the false detection width difference metric is calculated based on the width ratio between the target image to be determined and the false detection matching image. The method includes:

[0101] Formula (5)

[0102] in, This represents a measure of the false detection width difference. represents the width ratio between the target image to be judged and the falsely detected matching image; ln represents the natural logarithm function.

[0103] (2) The false detection height difference measure is calculated based on the target height vector and the matching height vector.

[0104] In this embodiment, the height ratio between the target image to be determined and the falsely detected matching image is calculated based on the target height vector of the target image to be determined and the matching height vector of the falsely detected matching image. The method includes:

[0105] Formula (VI)

[0106] in, This represents the height ratio between the target image to be determined and the falsely detected matching image; Represents the target height vector; This indicates matching the height vector.

[0107] In this embodiment, the false detection height difference metric is calculated based on the height ratio between the target image to be determined and the false detection matching image. The method includes:

[0108] Formula (VII)

[0109] in, This indicates a measure of the difference in false positive rates. The height ratio of the target image to the false positive matching image is represented by ln; ln represents the natural logarithm function.

[0110] (3) Calculate the false detection width similarity score based on the false detection width difference metric, including the following methods:

[0111] Formula (8)

[0112] in, This represents the false positive width similarity score; This represents a measure of the false detection width difference. represents the size similarity tolerance factor; exp represents the natural exponential function.

[0113] (4) Calculate the false detection height similarity score based on the false detection height difference metric, including the following methods:

[0114] Formula (IX)

[0115] in, Indicates the false positive high similarity score; This indicates a measure of the difference in false positive rates. represents the size similarity tolerance factor; exp represents the natural exponential function.

[0116] (5) Calculate the false detection size similarity score based on the false detection width similarity score and the false detection height similarity score, in the following ways:

[0117] Formula (10)

[0118] in, This represents the false positive size similarity score; This represents the false positive width similarity score; This represents the high similarity score for false positives.

[0119] In this embodiment, the size similarity tolerance factor The false positive size similarity score is 0.25. The value range is 0-1.0.

[0120] S403: A final target similarity score is calculated based on the human feature false detection similarity score, the false detection feature similarity score and the false detection size similarity score. The method comprises:

[0121] ; Formula (11)

[0122] wherein, represents the final target similarity score; represents the weight of false detection library similarity score; represents the false detection feature similarity score; represents the weight of human feature factor; represents the human feature false detection similarity score; represents the weight of size similarity score; represents the false detection size similarity score.

[0123] In this embodiment, the weight of false detection library similarity score is 0.5, the weight of human feature factor is 0.2, and the weight of size similarity score is 0.3.

[0124] S404: An alarm determination result is generated based on the final target similarity score and a preset similarity score threshold.

[0125] In this embodiment, a higher final target similarity score indicates a higher probability of false detection; a lower final target similarity score indicates a lower probability of false detection, and the detected target is more likely to be a real human.

[0126] In this embodiment, the final target similarity score is compared with the preset similarity score threshold M_shreshold. If the final target similarity score is greater than or equal to the preset similarity score threshold, that is ≥M_shreshold, it is determined that the current detection target is not a human body and is a false detection, and a corresponding alarm determination result is generated, where the alarm determination result is no alarm, and dangerous industrial equipment maintains normal operation. If the final target similarity score is less than the preset similarity score threshold, that is <M_shreshold, it is determined that the current detection target is a human body, and a corresponding alarm determination result is generated, where the alarm determination result is outputting an alarm, triggering shutdown, and controlling the dangerous industrial equipment to stop operation, so as to realize safety protection for on-site personnel.

[0127] In this embodiment, the method further includes: updating the false detection image library in real time, which includes: if the final target similarity score is greater than or equal to a preset similarity score threshold, then determining that the current detected target is a non-human false detection target, and adding the corresponding image to be identified to the false detection image library to expand the capacity of the false detection sample library. By updating the false detection image library in real time, the detected target in the image to be identified will not be falsely identified as a human body in the next target recognition process, thereby continuously improving the false detection filtering accuracy and robustness of the system.

[0128] It is worth noting that the human false detection filtering method based on multi-dimensional comparison of the present invention has the following advantages:

[0129] (1) Layered judgment design to improve detection efficiency and judgment accuracy: First, the target confidence result is compared with the preset false detection library execution threshold. When the target confidence result is greater than or equal to the preset false detection library execution threshold, an alarm is directly determined without entering the subsequent comparison process, thus improving detection efficiency. When the target confidence result is less than the preset false detection library execution threshold, based on the false detection filtering algorithm and using the pre-built false detection image library, a multi-dimensional judgment is performed, including human feature factor dual threshold screening, false detection library feature similarity comparison, and size similarity comparison. This not only avoids the missed detection of real human targets under occlusion and other conditions, but also effectively filters out non-human false detection targets with similar appearance, reducing the false detection rate and the missed detection rate.

[0130] (2) Reduce false detection rate and improve judgment reliability: By using the human feature false detection similarity score, the possibility of false detection is judged from the dimension of human feature similarity. By using the false detection feature similarity score and the false detection size similarity score, the images are compared with each non-human false detection sample image in the false detection image library from the dimensions of false detection image feature similarity and size similarity. Only non-human targets that are highly matched with the features and size of the false detection library samples are filtered and fused to obtain the final target similarity score. Real human targets are not blocked, thus realizing the distinction between non-human false detection targets and real human targets, reducing the false detection rate, ensuring the accuracy of false detection filtering, and improving the reliability and scene adaptability of alarm judgment results.

[0131] (3) Human feature dual threshold screening to reduce invalid comparisons: When the target confidence result is less than the preset false detection library execution threshold, the human feature factor to be judged is first compared with the preset human feature factor high threshold and preset human feature factor low threshold. This can quickly distinguish highly credible human, highly credible non-human and fuzzy targets. Only fuzzy targets whose human feature factor to be judged is greater than the preset human feature factor low threshold and less than the preset human feature factor high threshold are subjected to subsequent false detection library feature comparison. This can quickly filter a large number of clear targets, reduce invalid feature comparison and calculation operations, further improve the algorithm running efficiency, and avoid misjudgment of clear targets, reducing the risk of missed detection and false detection.

[0132] (4) Reduce training costs: False detection filtering can be achieved by comparing the preset threshold with the false detection image library. There is no need to rely on massive labeled data to expand the training set, and there is no need to frequently retrain the model. This reduces the cost of data collection, labeling and model training. The real-time update of the false detection image library can adapt to new scenarios and new interference targets, and achieve rapid and stable detection of people in dangerous areas.

[0133] Figure 7 This is a schematic block diagram of a human false detection filtering system based on multi-dimensional comparison provided in an embodiment of this application. Figure 7 As shown, this human false detection filtering system 700 based on multi-dimensional comparison:

[0134] The image acquisition module 701 is used to acquire images of the preset danger zone to be identified.

[0135] The target recognition module 702 is used to perform a target recognition operation based on the image to be recognized, so as to generate a target determination result; wherein the target determination result includes a target confidence result.

[0136] The false detection filtering module 703 is used to generate an alarm judgment result based on the target determination result and a false detection filtering algorithm and a preset false detection library execution threshold. The method includes: if the target confidence result is greater than or equal to the preset false detection library execution threshold, then the alarm judgment result is determined to be an output alarm; if the target confidence result is less than the preset false detection library execution threshold, then a multi-dimensional similarity comparison operation is performed based on the false detection filtering algorithm and using a pre-built false detection image library to generate an alarm judgment result.

[0137] It should be understood that the specific process of each module performing the above-mentioned steps has been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

[0138] It should also be understood that the module division in the embodiments of this application is illustrative and only represents a logical functional division; in actual implementation, there may be other division methods. Furthermore, the functional modules in the various embodiments of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0139] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method as described in any of the embodiments above.

[0140] As used in this specification, the terms "component," "module," "system," etc., are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, and / or a computer. As illustrated, applications running on computing devices and computing devices can both be components. One or more components may reside in a process and / or an execution thread, and components may be located on a single computer and / or distributed among two or more computers. Furthermore, these components can be executed from various computer-readable media on which various data structures are stored. Components can communicate, for example, via local and / or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, and / or a network, such as the Internet interacting with other systems via signals).

[0141] Those skilled in the art will recognize that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0142] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0143] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0144] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0145] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0146] In the above embodiments, the functions of each functional unit can be implemented entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. A computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0147] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] In summary, this application provides a method, system, and medium for human false detection filtering based on multi-dimensional comparison. First, it compares the target confidence result with a preset false detection library execution threshold. When the target confidence result is greater than or equal to the preset false detection library execution threshold, an alarm is directly triggered without proceeding to subsequent comparison processes, thus improving detection efficiency. When the target confidence result is less than the preset false detection library execution threshold, based on the false detection filtering algorithm and utilizing a pre-built false detection image library, multi-dimensional judgment is performed using dual-threshold screening of human feature factors, false detection library feature similarity comparison, and size similarity comparison. This eliminates the need to rely on massive amounts of labeled data to expand the training set, reducing the cost of data collection, labeling, and model training, and lowering the false detection and false negative rates. It also improves the reliability and scenario adaptability of the alarm judgment results, overcoming the shortcomings of insufficient generalization and fixed categories inherent in simple classification networks, as well as the inflexibility of region-based masking filtering for false detection. Therefore, this application effectively overcomes the various shortcomings of existing technologies and has high industrial application value.

[0149] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for filtering false positives in human bodies based on multi-dimensional comparison, characterized in that, include: Collect images of the pre-defined hazardous areas to be identified; Based on the image to be identified, a target recognition operation is performed to generate a target determination result; wherein, the target determination result includes a target confidence result; Based on the target determination result, and using a false detection filtering algorithm and a preset false detection library execution threshold, an alarm determination result is generated; the method includes: If the target confidence result is greater than or equal to the preset false detection library execution threshold, then the alarm determination result is determined to be an output alarm; If the target confidence score is less than the preset false detection library execution threshold, then based on the false detection filtering algorithm and using a pre-built false detection image library, a multi-dimensional similarity comparison operation is performed to generate an alarm judgment result; the method includes: A cropping operation is performed on the image to be identified to generate a target image to be judged; Based on the human feature extraction network, inference operations are performed on the target image to be judged to generate human feature factors to be judged. Based on the human feature factors to be determined, and using preset high thresholds for human feature factors, preset low thresholds for human feature factors, and the false detection image library, a multi-dimensional similarity comparison operation is performed to generate an alarm determination result; the method includes: If the human feature factor to be determined is greater than or equal to the preset high threshold of human feature factor, then the alarm determination result is determined to be an output alarm. If the human feature factor to be determined is less than the preset high threshold of human feature factor, then determine whether the human feature factor to be determined is less than or equal to the preset low threshold of human feature factor. If the human characteristic factor to be determined is less than or equal to the preset low threshold of human characteristic factor, then the alarm determination result is determined to be no alarm. If the human feature factor to be determined is greater than the preset low threshold of human feature factor and less than the preset high threshold of human feature factor, then a multi-dimensional similarity comparison operation is performed based on the false detection image library to generate an alarm judgment result. The false detection image library includes multiple non-human false detection sample images; wherein, based on the false detection image library, a multi-dimensional similarity comparison operation is performed to generate an alarm judgment result, including: Based on the human feature factors to be determined, perform a human feature similarity comparison operation to calculate the human feature false detection similarity score. Based on the target image to be determined and each non-human false detection sample image in the false detection image library, perform false detection feature similarity comparison operation and false detection size similarity comparison operation to calculate the false detection feature similarity score and the false detection size similarity score. The final target similarity score is calculated based on the false detection similarity score of human body features, the false detection feature similarity score, and the false detection size similarity score. Based on the final target similarity score and a preset similarity score threshold, an alarm determination result is generated.

2. The human false detection filtering method based on multi-dimensional comparison according to claim 1, characterized in that, Based on the target image to be determined and each non-human false detection sample image in the false detection image library, the methods for performing false detection feature similarity comparison operation and false detection size similarity comparison operation to calculate the false detection feature similarity score and false detection size similarity score include: Based on the false detection library model, a first feature extraction operation is performed on the target image to be determined to generate a target feature vector and a target size vector; Based on the false detection library model, a second feature extraction operation is performed on each of the aforementioned non-human false detection sample images to generate multiple false detection feature vectors and multiple false detection size vectors; wherein, each non-human false detection sample image corresponds to one false detection feature vector and one false detection size vector; Based on the target feature vector and each of the false detection feature vectors, a false detection feature similarity comparison operation is performed to lock the false detection matching image and the optimal matching size vector, and the false detection feature similarity score is calculated. Based on the target size vector and the optimal matching size vector, a false detection size similarity comparison operation is performed to calculate the false detection size similarity score.

3. The human false detection filtering method based on multi-dimensional comparison according to claim 2, characterized in that, The method for performing a false detection feature similarity comparison operation based on the target feature vector and each of the false detection feature vectors to lock the false detection matching image and the optimal matching size vector, and to calculate the false detection feature similarity score includes: Based on the target feature vector and each of the false detection feature vectors, a false detection feature similarity comparison operation is performed to calculate the Euclidean distance between the target feature vector and each of the false detection feature vectors, and the Euclidean distance with the smallest value is taken as the optimal matching Euclidean distance. In the false detection image library, the non-human false detection sample image corresponding to the optimal matching Euclidean distance is locked as the false detection matching image, and the false detection size vector of the non-human false detection sample image corresponding to the optimal matching Euclidean distance is used as the optimal matching size vector. The similarity score of false detection features is calculated based on the optimal matching Euclidean distance.

4. The human false detection filtering method based on multi-dimensional comparison according to claim 2, characterized in that, The target size vector includes a target width vector and a target height vector; the optimal matching size vector includes a matching width vector and a matching height vector; wherein, based on the target size vector and the optimal matching size vector, the method for performing a false detection size similarity comparison operation to calculate the false detection size similarity score includes: The false detection width difference metric is calculated based on the target width vector and the matching width vector. Based on the target height vector and the matching height vector, a false detection height difference metric is calculated. The false detection width similarity score is calculated based on the false detection width difference metric. The false detection height similarity score is calculated based on the false detection height difference metric. The false detection size similarity score is calculated based on the false detection width similarity score and the false detection height similarity score.

5. The human body false detection filtering method based on multi-dimensional comparison according to claim 1, characterized in that, The final target similarity score can be calculated based on the human body feature false detection similarity score, the false detection feature similarity score, and the false detection size similarity score, including the following methods: ； in, This represents the final target similarity score; Indicates the weight of the similarity score for the false detection library; This represents the similarity score of false positive features; Indicates the weight of human characteristic factors; This represents the similarity score for false positives of human features. Indicates the weight of the size similarity score; This represents the false positive size similarity score.

6. A human false detection filtering system based on multi-dimensional comparison, characterized in that, include: The image acquisition module is used to acquire images of the preset danger zones to be identified. The target recognition module is used to perform a target recognition operation based on the image to be recognized, so as to generate a target determination result; wherein, the target determination result includes a target confidence result; The false detection filtering module is used to generate an alarm judgment result based on the target determination result, a false detection filtering algorithm, a preset false detection library, and an execution threshold; the method includes: If the target confidence result is greater than or equal to the preset false detection library execution threshold, then the alarm determination result is determined to be an output alarm; If the target confidence score is less than the preset false detection library execution threshold, then based on the false detection filtering algorithm and using a pre-built false detection image library, a multi-dimensional similarity comparison operation is performed to generate an alarm judgment result; the method includes: A cropping operation is performed on the image to be identified to generate a target image to be judged; Based on the human feature extraction network, inference operations are performed on the target image to be judged to generate human feature factors to be judged. Based on the human feature factors to be determined, and using preset high thresholds for human feature factors, preset low thresholds for human feature factors, and the false detection image library, a multi-dimensional similarity comparison operation is performed to generate an alarm determination result; the method includes: If the human feature factor to be determined is greater than or equal to the preset high threshold of human feature factor, then the alarm determination result is determined to be an output alarm. If the human feature factor to be determined is less than the preset high threshold of human feature factor, then determine whether the human feature factor to be determined is less than or equal to the preset low threshold of human feature factor. If the human characteristic factor to be determined is less than or equal to the preset low threshold of human characteristic factor, then the alarm determination result is determined to be no alarm. If the human feature factor to be determined is greater than the preset low threshold of human feature factor and less than the preset high threshold of human feature factor, then a multi-dimensional similarity comparison operation is performed based on the false detection image library to generate an alarm judgment result. The false detection image library includes multiple non-human false detection sample images; wherein, based on the false detection image library, a multi-dimensional similarity comparison operation is performed to generate an alarm judgment result, including: Based on the human feature factors to be determined, perform a human feature similarity comparison operation to calculate the human feature false detection similarity score. Based on the target image to be determined and each non-human false detection sample image in the false detection image library, perform false detection feature similarity comparison operation and false detection size similarity comparison operation to calculate the false detection feature similarity score and the false detection size similarity score. The final target similarity score is calculated based on the false detection similarity score of human body features, the false detection feature similarity score, and the false detection size similarity score. Based on the final target similarity score and a preset similarity score threshold, an alarm determination result is generated.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the human false detection filtering method based on multi-dimensional comparison as described in any one of claims 1 to 5.

Images