Personnel protection wearing identification method, device, equipment, medium and program product

Abstract

The invention relates to the technical field of image recognition, and discloses a personnel protection wearing recognition method, device and equipment, a medium and a program product, the personnel protection wearing recognition method comprises the steps that one or more video streams of a target area are continuously acquired, and the target area is an area where a preset number of camera devices perform video acquisition; performing feature extraction processing on a first target image in the video stream, and determining a plurality of detection results of the current human body target according to features extracted from the first target image; performing primary matching on a first result in the plurality of detection results and an existing tracking trajectory, and if the primary matching result is that the primary matching fails, performing secondary matching on a second result in the plurality of detection results and the existing tracking trajectory; and identifying a second target image in the video stream corresponding to a re-matching result to judge whether the human body target correctly wears the protective equipment or not. The recognition accuracy of the human body target can be improved.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a method, device, equipment, medium and program product for identifying personnel protective wear. Background Art

[0002] In the fields of medicine, chemical industry, biopharmaceuticals, nuclear energy and special working environments, the correct wearing of personal protective equipment has become a key link in preventing occupational exposure and ensuring personnel safety. Personal protective equipment generally includes positive pressure hoods, protective masks, protective clothing, etc. The correct wearing of protective equipment is crucial to prevent the invasion of harmful substances and protect the respiratory system and physical safety of personnel.

[0003] The related technology collects images from a camera to identify whether personnel are wearing protective equipment correctly. However, the scenarios in special working environments are relatively complex. There are often multiple personnel, and the appearance of the personnel is highly similar. There are also interference factors such as various types of working equipment and facilities. Personnel are frequently obscured, which increases the difficulty in identifying the protective equipment worn by personnel and the recognition accuracy is relatively low. Summary of the invention

[0004] In view of this, the present invention provides a method, device, equipment, medium and program product for identifying personnel protective wear to solve the problem of relatively low recognition accuracy of existing recognition methods.

[0005] In a first aspect, the present invention provides a method for identifying protective wear of personnel, comprising continuously acquiring one or more video streams of a target area, wherein the target area is an area where a preset number of cameras perform video acquisition; performing feature extraction processing on a first target image in the video stream, and determining multiple detection results of a current human target based on features extracted from the first target image; performing an initial match on a first result among the multiple detection results with an existing tracking trajectory, and if the result of the initial match is a failure of the initial match, then re-matching a second result among the multiple detection results with the existing tracking trajectory; wherein the target detection score of the second result is less than the target detection score of the first result; and identifying the second target image in the video stream corresponding to the re-matched result to determine whether the human target is correctly wearing protective equipment.

[0006] Beneficial effect: In the method for identifying protective wear for personnel provided by the present invention, a first result with a relatively high score is first used for matching. If the initial matching fails, a second result with a relatively low score is used for matching, and the second target image corresponding to the successfully matched second result is identified. On the one hand, first using the first result for matching can improve the accuracy of matching and reduce the risk of false matching. On the other hand, when a high-scoring detection result fails, using a low-scoring detection result for re-matching is a safeguard measure that can still realize the identification function of protective wear for personnel when the initial matching fails. It can improve the matching success rate of tracking trajectories and detection results and avoid missing matches, thereby effectively tracking personnel who need to be monitored and identified, so as to facilitate the subsequent identification of protective wear for personnel. Therefore, the present invention can improve the tracking performance of human targets and more accurately and comprehensively identify whether personnel in the target area are wearing protective equipment correctly.

[0007] In an optional embodiment, the initial matching of the first result among the multiple detection results with the existing tracking trajectory includes: dividing the multiple detection results into the first result and the second result according to the target detection scores of the multiple detection results; wherein the target detection score of the second result is less than or equal to a first threshold, and the target detection score of the first result is greater than the first threshold; and the first result is initially matched with each currently stored tracking trajectory to obtain an initial matching result; wherein the initial matching result includes an initial matching success and an initial matching failure, wherein the initial matching success indicates that the currently stored tracking trajectories are all successfully matched with the first result, and the initial matching failure indicates that at least one currently stored tracking trajectory is not successfully matched with the first result.

[0008] Beneficial effect: By setting the first threshold, the detection result can be compared with the first threshold, so as to efficiently divide the first result and the second result. By dividing the result of the initial matching into two situations, different processing methods can be adopted according to different situations in the subsequent process, and the tracking trajectory that failed the initial matching can be matched again.

[0009] In an optional embodiment, the re-matching of the second result among the multiple detection results with the existing tracking trajectory includes: re-matching the at least one tracking trajectory that has not been successfully matched with the first result with the second result to obtain a re-matched result; determining the second target image in the video stream corresponding to the tracking trajectory that has been successfully matched with the second result; wherein the re-matching result includes a re-matching success and a re-matching failure, the re-matching success indicates that the second result is successfully matched with the tracking trajectory, and the re-matching failure indicates that the second result is not successfully matched with the tracking trajectory.

[0010] Beneficial effect: Using the second result for re-matching can successfully achieve matching of detection results and tracking trajectories when the initial matching fails. After the re-matching is successful, the second target image is extracted according to the second result, so that the protective wear of the personnel can be identified later according to the second target image.

[0011] In an optional embodiment, the method further includes: marking the tracking trajectory that has not been successfully matched with the second result within a first preset time period as a lost trajectory; and judging the lost trajectory after a second preset time period after the tracking trajectory is marked as the lost trajectory to determine whether to delete the lost trajectory from the existing tracking trajectory.

[0012] Beneficial effects: On the one hand, deleting the lost track can reduce the number of stored tracking tracks and improve the recognition speed; on the other hand, the unmatched tracking tracks are still retained for a period of time and continue to be matched, so that the present invention has the ability to track temporarily disappeared human targets, which helps to ensure the stable tracking performance of the present invention in complex scenes.

[0013] In an optional implementation, the method further includes performing reliability evaluation on the tracking trajectories that fail to match the second result, and deleting the tracking trajectories evaluated as unreliable from the existing tracking trajectories.

[0014] Beneficial effect: By evaluating the reliability of unsuccessfully matched tracking trajectories, unnecessary tracking trajectories can be identified and eliminated in advance, thereby more efficiently reducing the number of stored tracking trajectories and improving the recognition speed.

[0015] In an optional implementation, after identifying the second target image in the video stream corresponding to the re-matching result, the method further includes: generating an alarm message if the human target who is not wearing protective equipment properly is identified.

[0016] Beneficial effect: Alarm information is used to remind management personnel so that personnel who do not wear protective equipment properly can be dealt with in a timely manner to eliminate safety hazards.

[0017] In a second aspect, the present invention provides a device for identifying protective wear for personnel, and the device for identifying protective wear for personnel includes a video acquisition module, a target recognition module, a personnel tracking module and a wear recognition module; wherein the video acquisition module is used to continuously acquire one or more video streams of a target area, and the target area is an area where a preset number of cameras perform video acquisition; the target recognition module is used to perform feature extraction processing on a first target image in the video stream, and to determine multiple detection results of a current human target based on features extracted from the first target image; the personnel tracking module is used to initially match a first result among the multiple detection results with an existing tracking trajectory, and to re-match a second result among the multiple detection results with the existing tracking trajectory based on the result of the initial matching being a failure of the initial matching; wherein the target detection score of the second result is less than the target detection score of the first result; the wear recognition module is used to identify the second target image in the video stream corresponding to the re-matching result to determine whether the human target is correctly wearing protective equipment.

[0018] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the method for identifying personal protective wear of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method for identifying personal protective wear according to the first aspect or any corresponding embodiment thereof.

[0020] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, which are used to enable a computer to execute the method for identifying personal protective wear according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related technologies, the drawings required for use in the specific embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 is a schematic flow chart of a method for identifying protective clothing for personnel according to an embodiment of the present invention; Figure 2is a flow chart of another method for identifying protective clothing for personnel according to an embodiment of the present invention; Figure 3 yes Figure 2 A schematic diagram of a specific flow chart of step S203 in the method for identifying personnel protective wear; Figure 4 yes Figure 2 A schematic diagram of a specific flow chart of step S204 in the method for identifying personnel protective wear; Figure 5 is a structural block diagram of a device for identifying personnel protective wear according to an embodiment of the present invention; Figure 6 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0024] In the fields of medicine, chemical industry, biopharmaceuticals, nuclear energy and special working environments, the correct wearing of personal protective equipment has become a key link in preventing occupational exposure and ensuring personnel safety.

[0025] At present, there are some protective equipment wearing monitoring solutions based on sensor technology in related technologies. These solutions usually indirectly determine whether the hood is worn by monitoring the air pressure changes inside the equipment or the wearer's breathing status. Although such technologies have improved the degree of automation of monitoring to a certain extent, they often rely on complex sensor networks and complex algorithm processing, which is not only costly, but also may cause unstable performance in extreme environments or specific scenarios (such as areas with severe electromagnetic interference).

[0026] The defects and reasons of protective equipment wearing monitoring solutions based on sensor technology are: 1) Unstable performance: The protective wear monitoring system based on sensor technology is easily interfered by external environmental factors, such as temperature, humidity, electromagnetic field, etc., resulting in false alarms or missed alarms, which affects the reliability and accuracy of the system.

[0027] 2) High cost: Complex sensor networks and data processing systems lead to higher overall costs, increase the burden on users, and limit their application in a wider range of scenarios.

[0028] 3) Inconvenient to use: Some systems require professional personnel to install and maintain, and require high operating skills from the wearer, which increases the difficulty of use and training costs.

[0029] 4) Inability to provide real-time feedback: Traditional sensor technology is often unable to provide real-time wear status information to managers or wearers, resulting in the inability to take timely countermeasures when abnormalities occur.

[0030] In other related technologies, computer equipment identifies whether personnel are wearing protective equipment correctly by capturing images from a camera device. However, in complex scenes, personnel are frequently obscured and the appearance of personnel is highly similar, making it difficult to effectively track personnel who need to be monitored and identified, resulting in reduced recognition accuracy.

[0031] To this end, an embodiment of the present invention provides a method for identifying protective clothing for personnel, by using a first result and a second result to perform two matches with the tracking trajectory, so as to achieve the effect of improving tracking performance and improving recognition accuracy.

[0032] According to an embodiment of the present invention, an embodiment of a method for identifying protective wear of personnel is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0033] In this embodiment, a method for identifying protective clothing worn by personnel is provided, which can be used in computer equipment. Figure 1 is a flow chart of a method for identifying personnel protective wear according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps: Step S101 , continuously acquiring one or more video streams of a target area, where the target area is an area where a preset number of cameras perform video acquisition.

[0034] The target area is a special working environment that requires safety control (such as an isolation ward, a pharmaceutical workshop, a chemical workshop, etc.). The camera device (such as a surveillance camera) is arranged in or around the target area to capture the image of the target area, and then send it to the computer device through a communication connection. The computer device can receive the signal from the surveillance camera to obtain a video stream. In order to monitor and identify the wearing of protective equipment by personnel in real time, in theory, the camera device continuously and uninterruptedly captures the target area and sends the image to the computer device in real time, so that the computer device can continuously obtain the video stream.

[0035] The identification method of personnel protective wear of the embodiment of the present invention supports simultaneous input and parallel analysis of multiple video streams. For example, for a target area with a relatively simple scene and a small range, a single camera device can clearly and comprehensively capture all personnel to be monitored in the target area. At this time, only one video stream of the target area can be obtained from a single camera device; while for a target area with a complex scene and a wide range, a single camera device is difficult to fully cover the target area, so it is required to obtain multiple video streams. At this time, multiple cameras can be set at different positions and angles, and each camera device generates one video stream to fully cover the target area, fully capture all personnel to be monitored in the target area, and also facilitate cross-analysis of whether the personnel are wearing protective equipment correctly.

[0036] In addition, it is understandable that the video stream is transmitted after being encoded, so the computer device needs to decode the video stream when obtaining the video stream. For example, a video decoding library (FFmpeg) that supports multiple video encoding formats can be used to implement video stream decoding and frame extraction.

[0037] Step S102, performing feature extraction processing on the first target image in the video stream, and determining multiple detection results of the current human target according to the features extracted from the first target image.

[0038] It is understandable that the first target image comes from the video frame extracted after decoding the video stream, and the first target image is presented as multiple continuous / non-continuous images that can be arranged in chronological order. The current human target refers to the image information of a specific person required for the current protective equipment identification. For example, if the protective equipment currently to be identified is a mask or a face shield, the current human target is the head or upper body of the person to be identified. If the protective equipment currently to be identified is protective clothing, the current human target is the whole body of the person to be identified.

[0039] In addition to the human targets to be monitored and identified by the present invention, there are also various types of operating equipment and facilities in the target area, and sometimes there may be obstructions such as light, smoke, and dust. All of this information will be collected by the camera device, so there is redundant information in the video stream received by the computer device, which is not suitable for direct analysis and identification.

[0040] Accurately identifying all persons in the first target image (including potential human target objects) is a prerequisite for subsequent tracking and detection of wearing conditions, which can avoid erroneous processing of irrelevant objects. By obtaining the first target image from the video frame in the video stream, and then extracting features based on the first target image, it is possible to distinguish persons and non-personnel from the first target image based on the unique features of the persons, and then determine all the distinguished persons as possible detection results of the current human target.

[0041] Step S103, performing an initial match between a first result among the multiple detection results and the existing tracking trajectory. If the result of the initial match is a failure, performing a second match between the multiple detection results and the existing tracking trajectory again; wherein the target detection score of the second result is less than the target detection score of the first result.

[0042] A tracking trajectory includes the detection results of a specific person in each historical first target image (that is, the first target image that has been matched), which is used to represent the historical action route of the specific person. Generally speaking, there will be multiple people who need to be monitored and identified in the target area at the same time, which will generate multiple existing tracking trajectories. Therefore, it is necessary to match each tracking trajectory with the multiple detection results generated by the current first target image to update the tracking trajectory.

[0043] In addition to the personnel to be monitored and identified by the present invention, there are often other personnel who do not need to be monitored and identified in the target area, and there may be multiple types of personnel in the target area. Different types of personnel may have different requirements for wearing protective equipment, resulting in differentiated human target requirements.

[0044] For example, in a medical working environment, personnel may be divided into patients and medical staff. Medical staff need to wear protective equipment correctly, but patients are not required to wear such equipment. Another example is in a chemical working environment, personnel may be divided into close-range workers close to high-risk equipment and long-range workers relatively far away from high-risk equipment. Close-range workers need a higher level of protection (such as wearing full-body protective clothing), while long-range workers need a relatively lower level of protection (for example, they only need to wear masks and face shields for respiratory protection). The two types of personnel have different human targets.

[0045] Therefore, accurately locking the human target among people is one of the keys to avoid misjudgment. However, since the human target may be blocked by other people or objects in the target area, and the appearance similarity between the human target and other people may be high (for example, when they wear protective clothing), some target tracking algorithms in related technologies (such as Kalman filtering algorithm and Hungarian algorithm) will have a sharp drop in tracking performance when the data quality is not ideal.

[0046] To this end, the present invention sets a target detection score, which is used to characterize the quality of the detection result (such as occlusion status, clarity status, etc.). First, the first result with a relatively high target detection score is used for matching, which can improve the accuracy of matching and reduce the risk of false matching. For the tracking trajectory that has not been successfully matched, the present invention then uses the second result to supplement the matching, and uses the second result to "cover the bottom line" to screen out the trackable trajectories from the tracking trajectories that have not been successfully matched. Even in the case of a low target detection score, the target can be effectively tracked. Even in complex scenes (such as dense crowds and frequent target occlusion), stable tracking can be achieved, thereby improving the robustness and reliability of the system and avoiding missed detection.

[0047] Step S104, identifying the second target image in the video stream corresponding to the re-matching result to determine whether the human target is wearing protective equipment correctly.

[0048] The second target image is an image obtained by cropping the first target image according to the second result. The second target image should exclude as much unnecessary information as possible from the first target image and completely contain the information of the human target to be identified. After the second result is successfully matched, the present invention successfully tracks the human target that cannot be tracked in the related art. Next, the second target image corresponding to the second result is extracted from the first target image, and the second target image can be used to determine whether the human target is wearing protective equipment correctly.

[0049] The method for identifying protective wear for personnel provided in this embodiment introduces a low-score tracking strategy. When a high-score detection result fails, a low-score detection result is used for re-matching. This is a safeguard measure that can still realize the identification function of protective wear for personnel when the initial matching fails. In other words, it has a "safety net" effect, can improve the tracking performance of human targets, thereby effectively tracking personnel who need to be monitored and identified, and more accurately and comprehensively identifying whether personnel in the target area are wearing protective equipment correctly.

[0050] In addition, the method for identifying personnel protective wear provided in this embodiment supports multi-channel video stream input and the identification of multiple protective equipment, and can be widely used in medical, chemical, biopharmaceutical, nuclear energy and special working environments and other fields, thereby improving the universality and application value of the technology.

[0051] In this embodiment, a method for identifying protective clothing worn by personnel is provided, which can be used in the above-mentioned computer device. Figure 2 is a flow chart of a method for identifying personnel protective wear according to an embodiment of the present invention, such as Figure 2 As shown, the process includes the following steps: Step S201, video stream acquisition: continuously acquire one or more video streams of the target area, where the target area is the area where a preset number of cameras are performing video acquisition. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0052] In some optional embodiments, the video stream can be adaptively decoded. Specifically, the real-time frames are divided into dynamic scenes and static scenes by comparing the changes in image features of previous and subsequent frames. When the picture is static or changes slowly, the position of the human target changes less, and the difficulty of matching and recognition is relatively low. At this time, a lower decoding rate and resolution can be selected to save resources and bandwidth. When the scene changes quickly or there are more people in the picture, the difficulty of matching and recognition increases, and a higher decoding rate and resolution are switched to.

[0053] Step S202, human detection: performing feature extraction processing on the first target image in the video stream, and determining multiple detection results of the current human target according to the features extracted from the first target image.

[0054] Specifically, refer to Figure 3 , the above step S202 may include: Step S2021, image preprocessing: preprocessing the first target image, wherein the preprocessing includes at least one of normalization, cropping and scaling.

[0055] Since the models and parameters of the camera devices are unpredictable, the original first target image obtained from the video stream may not be suitable for subsequent related operations, for example, the resolution of the first target image may not be appropriate, or the first target image may contain a picture outside the target area, etc. In particular, when video streams are obtained from multiple cameras at the same time, the video streams of different cameras may also have inconsistent parameters.

[0056] By preprocessing the original data, it can be ensured that the input data meets the requirements of the identification method of personnel protective wear, thereby improving the accuracy of identification.

[0057] In some optional embodiments, the preprocessing may further include adaptive light adjustment and noise filtering. The adaptive light adjustment dynamically adjusts the brightness value and contrast coefficient of the first target image by calculating the global brightness and contrast of the first target image to adapt to the first target image under different light conditions. The adaptive noise filtering dynamically suppresses noise and smoothes the first target image by calculating the signal-to-noise ratio of the global image to reduce the impact of noise on target detection.

[0058] Step S2022, feature extraction: extracting features of the first target image, wherein the features include at least one of color, texture, shape, and depth.

[0059] As described above, the first target image contains information about people and non-people (objects, light and shadow). There are significant differences between the features of people and non-people in terms of color, texture, shape, and depth. Targeted extraction of these feature information helps to better distinguish people and non-people from the first target image.

[0060] Optionally, in this step, a deep learning algorithm can be used to extract features from the first target image. The deep learning algorithm uses a convolutional neural network (CNN) to automatically learn features in the image, so that the required features can be fully and accurately extracted from the complex first target image. Although the use of traditional algorithms (such as sliding window method, manual feature method) instead of deep learning algorithms for personnel detection has a small amount of calculation and is simple to implement, it is not effective in complex environments (such as light changes, occlusion, etc.), and the recognition accuracy is low, which will have a greater negative impact on subsequent steps.

[0061] Exemplarily, the deep learning algorithm introduces a feature multi-scale fusion strategy when extracting features, which enhances the expressiveness of features and further improves the detection capability of the model. Specifically, the feature fusion strategy adopted is a top-down and bottom-up bidirectional fusion structure, and introduces contextual information. The top-down path retains high-level semantic information and passes it to lower layers, while the bottom-up path uses the precise positioning information of the lower layers to enhance the entire feature hierarchy. This innovative structure can more effectively utilize feature information at different levels, and the introduction of contextual information enables the model to learn richer feature representations, improving the accuracy and robustness of predictions.

[0062] Step S2023, identifying human body information and marking the position: generating a bounding box to mark the position of the human target based on the features.

[0063] The bounding box is one of the methods that can accurately mark the target location. Using the bounding box for marking helps to obtain better output results.

[0064] Step S2024, data post-processing: performing data post-processing on the bounding box.

[0065] Due to the complexity of the image and the diversity of features, a huge number of bounding boxes will be generated based on the features of the first target image, of which only a portion of the bounding boxes accurately mark the human targets in the first target image. Therefore, data post-processing is required to screen the bounding boxes.

[0066] In some optional implementations, the above step S2024 includes: Step a1: Filter the bounding boxes based on their confidence, and remove the bounding boxes whose confidence is less than a second threshold.

[0067] Step a2: performing non-maximum suppression on the bounding boxes, and eliminating the bounding boxes whose overlapping area with the bounding box with the highest confidence is greater than a third threshold.

[0068] Confidence filtering can remove a large number of low-confidence bounding boxes (that is, bounding boxes where people hardly fall within the marked range), and non-maximum suppression can filter out bounding boxes with better marking effects. Through confidence filtering and non-maximum suppression, unreasonable bounding box results can be effectively removed, the number of bounding boxes can be reduced, and the accuracy of marking can be improved. Exemplarily, the second threshold can be any value between 0.2 and 0.3, and the third threshold can be any value between 0.6 and 0.8.

[0069] Step S2025, result output: output the remaining bounding box (that is, the bounding box after data post-processing) as the detection result.

[0070] Step S203, personnel tracking: initially matching the first result among the multiple detection results with the existing tracking trajectory. If the result of the initial matching is a failure, matching the second result among the multiple detection results with the existing tracking trajectory again; wherein the target detection score of the second result is less than the target detection score of the first result.

[0071] Specifically, refer to Figure 4 , the above step S203 may include: Step S2031, obtaining the detection result.

[0072] Step S2032, detection result division: divide the multiple detection results into a first result and a second result according to the target detection scores of the multiple detection results; wherein the target detection score of the second result is less than or equal to the first threshold, and the target detection score of the first result is greater than the first threshold.

[0073] In some optional embodiments, the confidence in step S202 may be directly used as the target detection score, and illustratively, the first threshold may be 0.5. In some other optional embodiments, a scoring standard may be established separately for the target detection score, and the target detection score may be calculated according to the scoring standard.

[0074] Optionally, the target detection score adopts a dynamic first threshold to further improve the accuracy and stability of tracking. Exemplarily, the first threshold selected when identifying the current first target image can be adaptively adjusted based on the detection result evaluation and trajectory quality evaluation of the previous first target image.

[0075] Among them, the detection result evaluation refers to analyzing the detection score of the human target in the previous first target image. This score usually indicates the confidence that the human target is correctly detected. If the detection score is high, it means that the detection result is more reliable; if the detection score is low, it may mean that there is an error in the detection result. Track quality evaluation refers to evaluating the track quality by calculating the smoothness, continuity and other indicators of the tracking track. A high-quality track should be smooth and continuous without obvious jumps or jitters.

[0076] When the detection score is high and the trajectory quality is high, the first threshold can be appropriately lowered to classify more detection results as the first results; otherwise, the first threshold needs to be increased.

[0077] Step S2033, initial matching: performing initial matching on the first result and each currently stored tracking track respectively to obtain the initial matching result; wherein the initial matching result includes a successful match and an initial matching failure, wherein the initial matching success indicates that all currently stored tracking tracks are successfully matched with the first result, and the initial matching failure indicates that at least one currently stored tracking track is not successfully matched with the first result.

[0078] It is understandable that the number of tracking tracks depends on the total number of human targets that need to be monitored and identified in the target area. In most cases, there are multiple tracking tracks, so the first result needs to be matched with the tracking tracks one by one. Only when all tracking tracks are successfully matched with the first result will the result of the first successful match be returned. If there is a tracking track that is not successfully matched, it is necessary to continue to find the corresponding detection result for the tracking track.

[0079] Step S2034, re-matching: re-matching at least one tracking trajectory that has not been successfully matched with the first result with the second result to obtain a re-matching result; determining a second target image in the video stream corresponding to the tracking trajectory that has been successfully matched with the second result; wherein the re-matching result includes a re-matching success and a re-matching failure, wherein a re-matching success indicates that the second result is successfully matched with the tracking trajectory, and a re-matching failure indicates that the second result is not successfully matched with the tracking trajectory.

[0080] Step S2035, creating and retaining a track: Optionally, if the result of the re-match is a re-match failure, the tracking track that failed to match again is retained for a third preset time period (such as m frames), and then matched with the detection result of the new first target image (that is, the first target image after m frames); Optionally, if the result of the initial matching is a failure, and the target detection score is greater than a fourth threshold, a new tracking track is created for the first result.

[0081] There are two possible reasons for the failure of re-matching. One is that the person corresponding to the tracking trajectory has left the target area. The other is that the human target corresponding to the tracking trajectory is severely blocked by other people or objects. For the latter, since the people and objects in the target area are not stationary, it can be expected that the human target will reappear after a period of occlusion.

[0082] The present invention uses the unmatched tracking track to match again after the third preset time period, so that the present invention has the ability to track the temporarily disappeared human target (that is, to match the cross-frame human target), which helps to ensure the stable tracking performance of the present invention in complex scenes, and matching again after the first time interval can also reduce unnecessary matching processes and improve the operating efficiency of the present invention. Exemplarily, m can be any value between 10 frames and 50 frames.

[0083] In addition, by incorporating high-scoring and unsuccessfully matched detection results into the tracking trajectory, the present invention has the ability to track human targets newly appearing in the target area, which helps to achieve comprehensive tracking of human targets in the target area and improves the reliability of the present invention.

[0084] In some optional embodiments, the fourth threshold may be higher than the target detection score line when dividing the first result and the second result. For example, when the first threshold is 0.5, the fourth threshold is 0.6, so as to further eliminate the detection results with relatively low scores in the first result. In other optional embodiments, the fourth threshold may also be equal to the target detection score line when dividing the first result and the second result, that is, to create a tracking track for all first results that have not been successfully matched.

[0085] Step S2036, track management: marking the tracking track that has not been successfully matched with the second result within the first preset time period as a lost track; after the tracking track is marked as a lost track for a second preset time period, judging the lost track to determine whether to delete the lost track from the existing tracking tracks.

[0086] It is understandable that the duration of the first preset time period is greater than the duration of m frames, and the tracking trajectory that has not been successfully matched with the second result within the first preset time period also means that there have been multiple unsuccessful matches, which means that the reason for the failure to match again is more likely that the person corresponding to the tracking trajectory has left the target area, so there is no need to monitor and identify it. By removing unnecessary tracking trajectories, the total amount of tracking trajectories stored in the computer device can be reduced, the matching speed can be increased, and the identification method of personnel protective wear can be run efficiently. The value of the first preset time period can be equal to 0. At this time, the tracking trajectory is deleted after being marked as a lost track, or if other management operations need to be introduced for the lost track, the value of the first preset time period can be greater than 0, so that the lost track is retained for a period of time.

[0087] In some optional implementations, step S2036 further includes performing reliability evaluation on the tracking trajectories that have not been successfully matched with the second result, and deleting the tracking trajectories evaluated as unreliable from the currently stored tracking trajectories.

[0088] By using reliability assessment, we can screen out in advance the tracking trajectories that have not been marked as lost trajectories and the corresponding people have left the target area with a high probability, and delete them. This not only reduces the risk of accidentally deleting valid tracking trajectories, but also improves the overall performance and robustness of the system.

[0089] Step S2037, returning the result: returning the updated tracking trajectory as the existing tracking trajectory to step S2031, waiting to be matched with a new detection result (that is, the detection result generated by the next first target image).

[0090] It is understandable that, when returning, the retention time of the tracking track that failed to match again in the existing tracking track will also be recorded. Only after the retention time reaches the preset first time, the tracking track that failed to match again will participate in the matching process of the next first target image.

[0091] Step S204, personnel protective wear identification: identify the second target image in the video stream corresponding to the re-matching result to determine whether the human target is wearing protective equipment correctly.

[0092] In some optional embodiments, a deep learning algorithm is used to identify the second target image. The deep learning algorithm has been trained in advance to learn a large amount of labeled data and fully understands the characteristics of when a person wears protective equipment correctly and when he does not wear protective equipment correctly, so it can correctly identify when a person wears protective equipment in accordance with the regulations.

[0093] In some optional embodiments, during training and learning, not only common features of wearing protective equipment, such as basic image contours and color features, are taken into consideration, but also the matching features between facial contours and mask edges are introduced, and then combined with high-level semantic features after image convolution, thereby improving the learning effect of the deep learning algorithm.

[0094] Optionally, a dynamic adjustment mechanism for environmental factors (such as light changes and background complexity) can be introduced into the deep learning algorithm so that it can adapt to different working environments and maintain a high recognition rate.

[0095] The method for identifying personnel protective wear in the embodiment of the present invention has significant effects and advantages compared with traditional manual inspection and sensor-based monitoring solutions by introducing computer vision and deep learning technology: 1) Real-time and accuracy: The present invention captures and analyzes video streams in real time, accurately identifying the personnel in the picture and the protective equipment they are wearing. In the test environment, the system's recognition accuracy of the protective equipment worn by personnel is as high as over 98%, and the false alarm rate and missed alarm rate are both less than 2%.

[0096] 2) Environmental adaptability: The present invention is not affected by the external environment (such as temperature, humidity, and electromagnetic field), and can operate stably in various complex environments, thereby improving the reliability and stability of the system.

[0097] 3) Scalability: It supports multi-channel video stream input and recognition of various protective equipment, and can be flexibly configured and expanded according to actual needs.

[0098] 4) Cost control: By optimizing the algorithm and hardware design, the overall cost of the present invention is reduced, making the technical solution have broader application prospects in more fields and scenarios.

[0099] Step S205, alarm: if a human target not wearing protective equipment properly is identified, an alarm message is generated.

[0100] The method for identifying personnel protective wear provided in this embodiment uses alarm information to prompt management personnel and related personnel, so that personnel who do not wear protective equipment properly can be dealt with in a timely manner to eliminate safety hazards.

[0101] In some optional implementations, the alarm information includes an event screenshot and an event tag, and the event tag includes time information, location information, and event category. Providing event screenshots and event tags helps managers quickly locate personnel who do not wear protective equipment properly, and also facilitates archiving and management of events.

[0102] Specifically, in some optional embodiments, the alarm information includes time information, location information, screenshot data, event category ("failure to wear protective clothing in a standardized manner", "failure to wear protective masks in a standardized manner", "failure to wear positive pressure headgear in a standardized manner"). Exemplarily, the alarm information is uploaded in the following format: { timestamp: "xxxx:xx:xx", location: "test1", image:”test2”, event: "Not wearing protective clothing properly" } After the alarm information is uploaded, it will be synchronously displayed on the client page (such as the administrator's operating terminal), making it convenient for the administrator to view, record, and rectify the problem in a timely manner.

[0103] In the present embodiment, a device for identifying personnel protective wear is also provided, and the device is used to implement the above-mentioned embodiments and preferred embodiments, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware is also possible and conceived.

[0104] This embodiment provides a device for identifying protective clothing worn by personnel, such as Figure 5 As shown, including: The video acquisition module 501 is used to continuously acquire one or more video streams of a target area, where the target area is an area where a preset number of cameras perform video acquisition.

[0105] The target recognition module 502 is used to perform feature extraction processing on the first target image in the video stream, and to determine multiple detection results of the current human target according to the features extracted from the first target image.

[0106] The personnel tracking module 503 is used to initially match the first result among the multiple detection results with the existing tracking trajectory, and to re-match the second result among the multiple detection results with the existing tracking trajectory based on the result of the initial matching being a failure of the initial matching; wherein the target detection score of the second result is less than the target detection score of the first result.

[0107] The wearing recognition module 504 is used to recognize the second target image in the video stream corresponding to the re-matching result to determine whether the human target is wearing the protective equipment correctly.

[0108] In some optional implementations, the target identification module 502 includes: The image preprocessing unit is used to preprocess the first target image, wherein the preprocessing includes at least one of normalization, cropping and scaling.

[0109] The feature extraction unit is used to extract features of the first target image, wherein the features include at least one of color, texture, shape, and depth.

[0110] The position marking unit is used to generate a bounding box marking the position of the human target according to the features.

[0111] The data post-processing unit is used to perform data post-processing on the bounding box.

[0112] The result output unit is used to output the remaining bounding box (that is, the bounding box after data post-processing) as the detection result.

[0113] In some optional implementations, the personnel tracking module 503 includes: The result acquisition unit is used to acquire the detection result.

[0114] A result division unit is used to divide the multiple detection results into a first result and a second result according to the target detection scores of the multiple detection results; wherein the target detection score of the second result is less than or equal to the first threshold, and the target detection score of the first result is greater than the first threshold.

[0115] The initial matching unit is used to perform initial matching on the first result and each currently stored tracking track respectively to obtain the initial matching result; wherein the initial matching result includes a successful match and an initial matching failure, wherein the initial matching success indicates that the currently stored tracking tracks are all successfully matched with the first result, and the initial matching failure indicates that at least one currently stored tracking track is not successfully matched with the first result.

[0116] The re-matching unit is used to re-match at least one tracking trajectory that has not been successfully matched with the first result with the second result to obtain a re-matching result; determine the second target image in the video stream corresponding to the tracking trajectory that has been successfully matched with the second result; wherein the re-matching result includes a re-matching success and a re-matching failure, wherein a re-matching success indicates that the second result is successfully matched with the tracking trajectory, and a re-matching failure indicates that the second result is not successfully matched with the tracking trajectory.

[0117] A new creation and retention track unit is used to retain the tracking track that failed to match again for a third preset time period (such as m frames) when the result of the re-matching is a re-matching failure, and then match it with the detection result of the new first target image (that is, the first target image after m frames); when the result of the initial matching is a failure and the target detection score is greater than a fourth threshold, create a new tracking track for the first result.

[0118] The track management unit is used to mark the tracking track that has not been successfully matched with the second result within a first preset time period (that is, the tracking track that has not been successfully matched within a period of time) as a lost track; after the tracking track is marked as a lost track for a second preset time period, the lost track is judged to determine whether to delete the lost track from the existing tracking tracks.

[0119] The track returning unit is used to return the updated tracking track as the existing tracking track to the result obtaining unit, waiting for matching with the new detection result.

[0120] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0121] The identification device for personnel protective wear in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0122] The embodiment of the present invention also provides a computer device having the above Figure 6 Identification device for the protective clothing worn by the personnel shown.

[0123] See also Figure 6 , Figure 6 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 6 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 10 is taken as an example.

[0124] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0125] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0126] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0127] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0128] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0129] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0130] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0131] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for identifying personnel protective wear, characterized in that: The method comprises: Continuously acquiring one or more video streams of a target area, where the target area is an area where a preset number of cameras are performing video acquisition; Performing feature extraction processing on a first target image in the video stream, and determining multiple detection results of a current human target according to the features extracted from the first target image; Performing an initial match between a first result among the multiple detection results and an existing tracking track, and if the result of the initial match is that the initial match fails, performing a second match between the second result among the multiple detection results and the existing tracking track again; wherein the target detection score of the second result is less than the target detection score of the first result; The second target image in the video stream corresponding to the re-matching result is identified to determine whether the human target is wearing protective equipment correctly.

2. The method according to claim 1, characterized in that The initially matching the first result among the multiple detection results with the existing tracking trajectory includes: According to the target detection scores of the multiple detection results, the multiple detection results are divided into the first result and the second result; wherein the target detection score of the second result is less than or equal to a first threshold, and the target detection score of the first result is greater than the first threshold; Performing a primary match between the first result and each currently stored tracking trajectory to obtain a primary matching result; The result of the initial matching includes initial matching success and initial matching failure, wherein the initial matching success indicates that all currently stored tracking trajectories successfully match the first result, and the initial matching failure indicates that at least one currently stored tracking trajectory does not successfully match the first result.

3. The method according to claim 2, characterized in that The re-matching of the second result among the multiple detection results with the existing tracking trajectory includes: Rematching the at least one tracking track that is not successfully matched with the first result with the second result to obtain a rematched result; Determine a second target image in the video stream corresponding to the tracking trajectory that successfully matches the second result; The result of the re-matching includes a re-matching success and a re-matching failure, wherein the re-matching success indicates that the second result successfully matches the tracking trajectory, and the re-matching failure indicates that the second result does not successfully match the tracking trajectory.

4. The method according to any one of claims 1 to 3, characterized in that Also includes: Marking the tracking track that has not been successfully matched with the second result within a first preset time period as a lost track; After the tracking trajectory is marked as the lost-tracking trajectory for a second preset time period, the lost-tracking trajectory is judged to determine whether to delete the lost-tracking trajectory from the existing tracking trajectory.

5. The method according to any one of claims 1 to 3, characterized in that Also includes: Reliability evaluation is performed on the tracking trajectories that have not been successfully matched with the second result, and the tracking trajectories that are evaluated as unreliable are deleted from the existing tracking trajectories.

6. The method according to any one of claims 1 to 3, characterized in that After identifying the second target image in the video stream corresponding to the re-matching result, the method further includes: If the human target who does not wear protective equipment correctly is identified, an alarm message is generated.

7. A device for identifying personal protective equipment, characterized in that: The device comprises: A video acquisition module, used to continuously acquire one or more video streams of a target area, wherein the target area is an area where a preset number of cameras perform video acquisition; A target recognition module, used for performing feature extraction processing on a first target image in the video stream, and for determining multiple detection results of a current human target according to features extracted from the first target image; A personnel tracking module, used for initially matching a first result among the multiple detection results with an existing tracking track, and for re-matching a second result among the multiple detection results with the existing tracking track according to a result of the initial matching being a failure of the initial matching; wherein the target detection score of the second result is less than the target detection score of the first result; The wearing recognition module is used to recognize the second target image in the video stream corresponding to the re-matching result to determine whether the human target is wearing protective equipment correctly.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for identifying personal protective wear according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method for identifying personal protective wear according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions are used to cause a computer to execute the method for identifying personal protective wear according to any one of claims 1 to 6.

Images