A teaching equipment tracing system based on computer vision and course schedule cross-validation

Abstract

The application discloses a kind of based on computer vision and course table cross-validation teaching equipment traceability system, the present application is analyzed in real time to classroom monitoring video by computer vision identification module, and is combined with human posture estimation, equipment identification and personnel identity recognition technology, realize the automatic monitoring and record of teaching equipment use process, without teacher active registration or additional operation can complete equipment use information acquisition, effectively reduce the manpower investment of equipment management, improve equipment traceability efficiency and management automation level.The present application associates and verifies the computer vision identification result with the course table data, determines the actual user of teaching equipment through the double-checking mechanism of visual identification and course information, can effectively reduce the misjudgment situation caused by face recognition error, environmental interference or equipment misrecognition, improve the accuracy and reliability of teaching equipment user identification, so as to improve the accuracy of equipment responsibility identification.

Description

Technical Field

[0001] This invention belongs to the field of teaching equipment traceability, specifically, it relates to a teaching equipment traceability system based on cross-validation of computer vision and timetable. Background Technology

[0002] With the popularization of information technology in teaching, various teaching equipment is widely used in daily school teaching activities. For example, wireless microphones, laser pointers, portable megaphones, and teaching tablet terminals are usually purchased and centrally managed by the school and shared among different classrooms or teachers. While these teaching devices can effectively improve classroom teaching efficiency, the large number of devices, frequent turnover, and constantly changing users are gradually increasing the difficulty of equipment management.

[0003] In practical applications, situations frequently arise where teaching equipment is lost, damaged, or not returned as required. When equipment malfunctions, school administrators typically need to identify the last person to use the equipment and the details of its use in order to determine responsibility and handle subsequent matters. However, because most teaching equipment is used in an open manner and lacks an effective end-to-end tracking mechanism, tracing responsibility for equipment becomes difficult, posing a significant challenge to school equipment management.

[0004] Currently, schools primarily manage teaching equipment through paper-based registration, electronic spreadsheets, or equipment borrowing and returning management systems. These methods generally rely on teachers proactively registering borrowing and returning information, which is prone to omissions, errors, or inaccuracies due to human negligence, resulting in incomplete equipment usage records. Furthermore, when equipment is lost, relying solely on registration information makes it difficult to accurately identify the actual user, making responsibility determination challenging. In addition, manually verifying historical records consumes significant time and manpower, leading to low management efficiency.

[0005] To improve equipment management, some schools have begun using technologies such as RFID tags and QR codes to manage teaching equipment. While these technologies can identify equipment and manage borrowing and returning records, users still need to actively scan codes or swipe cards. They cannot automatically obtain information about the personnel involved in actual equipment use, nor can they accurately determine the actual user of the equipment at a specific time and place. Therefore, when equipment is lost or there is a dispute over liability, existing technologies still struggle to establish a complete, objective, and reliable chain of evidence regarding equipment use, failing to meet the practical needs of refined management and accountability for teaching equipment.

[0006] In view of this, the present invention is proposed. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a teaching equipment traceability system based on cross-validation of computer vision and timetable, which solves the problems mentioned in the background art.

[0008] To solve the above-mentioned technical problems, the basic concept of the technical solution adopted by the present invention is as follows: A teaching equipment traceability system based on cross-validation of computer vision and timetable includes: a computer vision recognition module for acquiring video stream data through a classroom monitoring camera and preprocessing the video stream data; using a deep learning target detection model to identify teaching equipment targets in the video images to obtain equipment category information, location information, and time information of the teaching equipment; and establishing the association between teaching equipment and users based on a human pose estimation algorithm, and identifying the person holding the equipment to obtain the corresponding teacher identity information. The timetable data processing module is used to establish data communication with the school's academic affairs system, obtain course arrangement information, teacher information, and classroom mapping relationships corresponding to the teaching scenario, and establish a timetable database. The cross-validation module receives teacher identity information, teaching equipment location information, and time information output by the computer vision recognition module, and performs correlation verification in conjunction with course arrangement information output by the timetable data processing module to determine the actual user of the target teaching equipment and generate a teaching equipment usage record. When the correlation verification result is abnormal, the module verifies course change information, substitute teaching information, and class rescheduling information, and updates the teaching equipment usage record according to the verification result, or generates an abnormal verification record and sends it to the management terminal for review. The cross-validation module is used to query the course schedule information in the timetable data based on the time information and classroom location information corresponding to the teaching equipment, and to match and verify the teacher identity information output by the computer vision recognition module with the teaching teacher information in the course schedule information; when the match is successful, a device usage association record is generated, and when the match fails, an abnormal verification record is generated.

[0009] The equipment feature library module is used to store and manage the equipment feature information of various teaching equipment, and to perform feature matching and assisted identification of teaching equipment based on at least one of the equipment appearance features, identification features, RFID features and behavioral features to determine the identity information of teaching equipment; at the same time, it receives the equipment identification information sent by the RFID reading device and combines the computer vision recognition results to perform fusion verification of the identity of teaching equipment.

[0010] The traceability report module is used to receive the verification results output by the cross-validation module, and construct a traceability information chain for the equipment based on the teaching equipment usage records, teacher identity information, timetable verification results, and abnormal event information; generate the equipment usage trajectory based on the location data and user data of the teaching equipment at different time points, and determine the circulation process and responsibility of the teaching equipment in combination with the equipment traceability information chain, thereby generating a traceability report for the teaching equipment.

[0011] Preferably, the device feature information includes at least one of appearance features, identification features, RFID features, and behavioral features.

[0012] Preferably, the deep learning target detection model in the computer vision recognition module is a YOLO series target detection model, which realizes the target detection of teaching equipment by recognizing the appearance features of the equipment in the video image; S21. Preprocess the acquired video stream images and input them into the YOLO object detection network for feature extraction; S22. Multi-scale feature extraction of target regions in video images is performed through a backbone network to obtain candidate target feature information of teaching equipment. S23. Based on the candidate target feature information, predict the bounding box of the teaching equipment target to obtain the position parameters of the teaching equipment target. The bounding box prediction formula is: as well as: in, It is the Sigmoid activation function. The x-coordinate of the center point of the prediction box. This represents the lateral offset predicted by the network. This represents the vertical offset predicted by the network. The x-coordinate of the top-left corner of the current grid cell. The ordinate of the top-left corner of the current grid cell. To predict the box width, To predict the bounding box height, The width of the prior bounding box (Anchor). The height of the prior bounding box (Anchor) The width scaling parameter is obtained from the network prediction. Here, e is the height scaling parameter predicted by the network, and e is a natural constant, approximately equal to 2.71828. S24. Calculate the target confidence score based on the bounding box prediction results to determine whether the teaching equipment target exists. The target confidence score calculation formula is as follows: The formula for calculating the Intersection over Union (IoU) is: ,in, This represents the area of ​​the intersection region between the predicted bounding box and the ground truth bounding box. Let Pr(Object) be the area of ​​the union region of the predicted bounding box and the ground truth bounding box, and let loU(pred,truth) be the intersection-union ratio of the predicted bounding box and the ground truth bounding box. S25. Iteratively train the YOLO object detection model based on the loss function to improve the object detection accuracy of teaching equipment. The loss function adopted is CIoU Loss, and its expression is: The bounding box loss function is: ,in, For the total loss function, To locate the loss weights, Weight the target confidence level loss. For classification loss weights, For bounding box loss, For target confidence loss, For classifying losses, This is the distance between the center point of the predicted bounding box and the center point of the ground truth bounding box. This is the diagonal length of the smallest bounding rectangle between the predicted bounding box and the ground truth bounding box. This parameter ensures consistency in aspect ratio between the predicted bounding box and the ground truth bounding box. The aspect ratio weighting coefficient; S26. Output the target detection results of the teaching equipment to obtain the equipment category information, location information and corresponding time information of the teaching equipment.

[0013] Preferably, the human pose estimation algorithm is the OpenPose algorithm, which determines the position of the hand region by detecting key points on the human body and establishes a correspondence between the teaching equipment and the user. Its steps include: S31. Perform human target detection on the acquired video images and input the human images into the OpenPose human pose estimation network; S32. Based on the OpenPose human pose estimation network, human key points are detected, and confidence maps corresponding to each human key point are generated. The calculation formula for the key point confidence map is as follows: ,in, Indicates the first The first personal physical goal The confidence values ​​corresponding to each key point; p represents the coordinates of any pixel in the image; Indicates the first The first personal physical goal The actual coordinates of each key point; This represents the squared Euclidean distance between a pixel and its corresponding keypoint. The standard deviation of the Gaussian distribution is used to control the diffusion range of the influence area of ​​key points. Represents an exponential function; S33. Generate Part Affinity Fields (PAF) vector fields based on the connection relationships between key points of the human body to establish the structural association of human limbs. The calculation formula for the Part Affinity Fields is as follows: ,in, Indicates the body limb area Corresponding PAF supervision information; This represents a human limb region consisting of two key points; p represents a pixel within the limb region. This represents the predicted vector of the PAF vector field at the corresponding pixel. and These represent the coordinates of key points at both ends of the limb; Indicates key points Pointing to key points The unit direction vector; S34. Obtain the position of the key point of the human wrist based on the human key point detection results, and obtain the center position of the teaching equipment in combination with the target detection results of the teaching equipment. S35. Calculate the Euclidean distance between the center position of the teaching equipment and the key point position of the human wrist. The distance calculation formula is as follows: ,in, The coordinates of the center position of the teaching equipment Coordinates of key points on the human wrist; S36. When the Euclidean distance satisfies the following correlation condition: Then it is determined that there is a possession relationship between the corresponding personnel and the teaching equipment, whereby... Here, d represents the distance threshold, indicating the distance between the teaching device and the wrist key point. S37. Based on the holding relationship, establish the correspondence between teaching equipment and users, and output the equipment-person association result.

[0014] Preferably, the device feature library module employs an adaptive weight allocation multi-feature fusion strategy to identify teaching devices, the multi-feature fusion strategy including: The feature weights are dynamically calculated based on the confidence levels corresponding to each device feature. The formula for calculating the feature weights is as follows: ,in, Indicates the first The weight value corresponding to each device feature is used to reflect the degree of contribution of that feature to the final recognition result; Indicates the first The confidence level of each device feature is used to characterize the reliability of the current feature recognition result; This represents the sum of confidence levels of all device features involved in the fusion, where Indicates the sequence numbers of all features involved in the fusion calculation; When the confidence level of the corresponding device feature is lower than the preset low confidence threshold, a reliability penalty is applied to the weight of the corresponding feature. The final identification result is obtained by fusing multiple device features using a weighted voting method. The fusion decision formula is as follows: ,in, This represents the final recognition result or comprehensive matching score after multi-feature fusion. This represents the weight value corresponding to the i-th device feature; Indicates the first The identification results, matching scores, or feature determination values ​​of each device feature; This indicates that the recognition results of all device features are weighted and summed according to their corresponding weights.

[0015] Preferably, the device feature library module is further configured to reset the corresponding feature weight to zero and renormalize the remaining feature weights when a single device feature fails; and to output unrecognizable information and trigger manual processing when all device features fail. The device feature library module also judges the recognition result based on a preset decision threshold. When the final recognition result is greater than the acceptance threshold, the recognition pass result is output. When the final recognition result is lower than the rejection threshold, the recognition failure result is output. When the final recognition result is between the acceptance threshold and the rejection threshold, the multi-feature fusion review flow is triggered.

[0016] Preferably, the computer vision recognition module is also used to adaptively optimize video images for low light, target occlusion, and multiple people in the same frame in teaching scenarios, so as to improve the accuracy of target recognition and personnel identification of teaching equipment.

[0017] The computer vision recognition module performs scene optimization processing for low-light conditions, target occlusion, and multiple people in the same frame in teaching scenarios. In low-light scenes, the computer vision recognition module enhances the brightness of the input image using the CLAHE adaptive histogram equalization algorithm, and the enhancement formula is as follows: , among which, among which, This represents the image after enhancement processing; I represents the original input image captured by the camera. This represents the contrast-limited adaptive histogram equalization processing function.

[0018] Next, the image reflection component is enhanced based on the Retinex enhancement algorithm, and the enhancement formula is as follows: ,in, Indicates the image in coordinates Enhancement results at the location; This represents the brightness information of the input image at the corresponding location; This represents the illumination estimation component generated by Gaussian filtering; and These represent the horizontal and vertical coordinates in the image, respectively. Simultaneously, infrared and visible light images are fused to improve the device's detection capabilities under low-light conditions; the fusion formula is as follows: ,in, This represents the merged image; Represents a visible light image; Represents an infrared image; This represents the fusion weighting coefficient, which ranges from 0 to... This represents the weight corresponding to the infrared image.

[0019] Preferably, in a target occlusion scenario, the computer vision recognition module generates an occlusion region mask using an occlusion detection model to identify the visible area of ​​the target. The occlusion detection formula is as follows: ,in, Indicates the masking area; Indicates the input image; This represents the occlusion detection model. Based on the occlusion region mask, features of the visible target region are extracted. The feature extraction formula is as follows: ,in, Indicates the characteristics of the visible area of ​​the target; Represents the feature extraction function; Indicates the input image; Indicates the masking area; Indicates the visible area mask; This represents element-wise multiplication. Simultaneously, the computer vision recognition module calculates the target matching score using a partial feature matching method, with the following scoring formula: ,in, This indicates the overall matching score of the target. Indicates the first The weights corresponding to each feature; The first term of the target to be identified One feature; This represents the first target in the device feature library. One feature; This represents the function for calculating feature similarity. Preferably, in a scenario with multiple people in the same frame, the computer vision recognition module sorts the detected faces according to their face area, prioritizing the recognition of larger face targets. Simultaneously, it determines the person holding the device based on the distance between the teaching equipment and the face location, using the following selection formula: ,in, This represents the final determined personnel holding the equipment; argmin represents the target corresponding to the minimum value. Indicates the coordinates of the center position of the face; Indicates the coordinates of the center position of the teaching equipment; The Euclidean distance between the two is represented, and the DeepSORT multi-object tracking algorithm is used to maintain the consistency of personnel IDs; the computer vision recognition module also matches target features based on cosine similarity, and the matching formula is: ,in, Indicates the similarity between feature vectors; This represents the feature vector of the target to be identified; This represents a reference feature vector in the feature library; Represents the dot product of two vectors; Representing vectors The modulus length; Representing vectors The length of the module.

[0020] Different recognition thresholds are used for different teaching scenarios, with a recognition threshold of 0.6 for normal scenarios, 0.55 for low-light scenarios, and 0.65 for occluded scenarios.

[0021] Preferably, the device feature library module employs an adaptive weight allocation multi-feature fusion strategy to identify teaching equipment, including: The feature weights are dynamically calculated based on the confidence levels corresponding to each device feature. The formula for calculating the feature weights is as follows: ,in, Indicates the first The weight values ​​corresponding to each device feature; Indicates the first Confidence level of each device feature; The first to participate in the integration The confidence level of each device feature is calculated; by normalizing the confidence level of each device feature, the contribution weight of different features in the final recognition result is dynamically allocated.

[0022] When the confidence level corresponding to the i-th device feature When the confidence level is below the preset low confidence threshold Tlow, a reliability penalty correction is applied to the corresponding feature weights so that the corrected feature weights satisfy the following: ,in, This represents the feature weights after reliability penalty; This represents the feature weights before penalty; 0.5 represents the weight decay coefficient; and when the confidence level of the corresponding feature is... Below the preset low confidence threshold When this happens, the weights of the features are attenuated to reduce the impact of low-reliability features on the final recognition result.

[0023] The fusion decision is based on the weights corresponding to the features of each device and the recognition results. A weighted voting method is used to obtain the final recognition result. The fusion decision formula is as follows: ,in, This represents the final recognition result after multi-feature fusion; Indicates the first The weights corresponding to each device feature; Indicates the first The recognition result or matching score corresponding to each device feature.

[0024] The fusion decision also adopts a cascaded decision method. When there is a single device feature that meets the preset high confidence condition, the recognition result of the corresponding device feature is directly used as the final recognition result. When any device feature fails, the corresponding feature weight is reset to zero, and the weights of the remaining device features are normalized; when all device features fail, an unrecognizable message is output and a manual processing procedure is triggered. The device feature library module also determines the final recognition result based on preset decision thresholds, where the acceptance threshold Taccept is 0.85 and the rejection threshold Treject is 0.50. When the final recognition result is greater than or equal to Taccept, a recognition pass result is output; when the final recognition result is less than or equal to Treject, a recognition failure result is output; when the final recognition result is between Treject and Taccept, a multi-feature fusion review process is triggered. When the i-th device feature fails, its corresponding weight satisfies: The weights corresponding to the remaining valid device features are then normalized to meet the following requirements. ,in, This represents the weights after renormalization; This represents the original weights after removing failure features; This represents the weight corresponding to the remaining valid device features; This represents the valid feature index. This formula is used to readjust the weight distribution of the remaining features after some features have become invalid. The device feature library module also determines the final identification result based on a preset decision threshold. When: Output the recognition result in real time; When Ffinal ≤ Treject, output a recognition failure result; When Treject < Ffinal < Taccept, the multi-feature fusion review process is triggered; where Taccept is the acceptance threshold, Treject is the rejection threshold, and Taccept is greater than Treject.

[0025] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all of the following advantages at the same time: (1) This invention uses a computer vision recognition module to analyze classroom monitoring videos in real time, and combines human posture estimation, equipment recognition and personnel identification technologies to realize automatic monitoring and recording of the use of teaching equipment. The equipment usage information can be collected without teachers having to actively register or perform additional operations, which effectively reduces the manpower input of equipment management and improves the efficiency of equipment traceability and the level of management automation.

[0026] (2) This invention associates and verifies computer vision recognition results with course schedule data. By using a dual verification mechanism of visual recognition and course information, the actual users of teaching equipment can be determined. This can effectively reduce misjudgments caused by facial recognition errors, environmental interference, or equipment misidentification, improve the accuracy and reliability of identifying users of teaching equipment, and thus improve the accuracy of equipment responsibility determination.

[0027] (3) This invention can automatically record the location information, user information, and course verification information of teaching equipment at different time points, and generate a complete equipment usage trajectory and traceability information chain, realizing data retention throughout the entire process of teaching equipment use. When equipment is lost, damaged, or there is a dispute over liability, relevant records can be quickly retrieved for analysis, providing objective, complete, and verifiable data evidence for equipment liability determination and event tracing.

[0028] (4) This invention makes full use of the school's existing surveillance camera system and academic affairs management system to obtain equipment usage information and course data. It can be deployed without large-scale hardware modifications to the existing teaching environment, thus reducing system construction and maintenance costs. At the same time, this invention is applicable to various teaching scenarios such as primary and secondary schools, universities, and training institutions, and has good versatility, economy, and promotion and application value.

[0029] (5) The present invention sets up a device feature library module and combines multiple device features such as device appearance features, identification features, RFID features and behavioral features for fusion identification. Even in complex teaching environments such as low light, partial obstruction or multiple people appearing at the same time, it can still improve the stability and accuracy of teaching device identification and enhance the system's adaptability and robustness in actual application scenarios.

[0030] (6) The present invention sets up an anomaly verification mechanism. When the visual recognition result is inconsistent with the timetable data, it can further combine the course change information, substitute teaching information and class rescheduling information for secondary verification and support manual review, thereby ensuring the accuracy of the judgment results of the equipment users in special teaching scenarios and improving the fault tolerance and practical application reliability of the system.

[0031] The specific embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0032] The accompanying drawings described below are merely some embodiments. Those skilled in the art can obtain other drawings based on these drawings without any creative effort. In the drawings: Figure 1 System overall architecture diagram.

[0033] Figure 2 : Flowchart of computer vision recognition.

[0034] Figure 3 Cross-validation algorithm flowchart.

[0035] Figure 4 : Schematic diagram of device feature recognition.

[0036] Figure 5 : A schematic diagram of the traceability report interface.

[0037] It should be noted that these accompanying drawings and textual descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art by referring to specific embodiments. Detailed Implementation

[0038] The invention will now be described in further detail with reference to the accompanying drawings.

[0039] Please see Figure 1-5 As shown, this embodiment provides a teaching equipment traceability system based on cross-validation of computer vision and timetables. It includes a computer vision recognition module for collecting video stream data through a classroom monitoring camera and preprocessing the video stream data; using a deep learning target detection model to identify teaching equipment targets in the video images to obtain equipment category information, location information, and time information; and establishing a correlation between teaching equipment and users based on a human pose estimation algorithm, and identifying the personnel holding the equipment to obtain corresponding teacher identity information. In this embodiment, the computer vision recognition module is deployed in the classroom surveillance camera and communicates with the backend server to monitor and recognize the use of teaching equipment in real time.

[0040] Specifically, the surveillance camera continuously collects video stream data from the classroom and sends the collected video images to the computer vision recognition module. The computer vision recognition module first uses the YOLOv8 object detection model to detect teaching equipment in the video images in real time, identifying teaching equipment targets such as handheld microphones, page turners, and tablet computers, and obtaining the position coordinates and time information of the corresponding devices.

[0041] After detecting the teaching equipment, the computer vision recognition module further uses the OpenPose human pose estimation algorithm to detect key points of the human body in the video image. Based on the spatial relationship between the location of the teaching equipment and the location of key points of the human hand, the correspondence between the teaching equipment and the user is established, thereby determining the target person currently holding the teaching equipment.

[0042] Subsequently, face detection and face recognition processing are performed on the identified target personnel, and the extracted facial features are compared with a pre-established teacher identity database to obtain the corresponding teacher's identity information, thereby realizing the association between teaching equipment and teacher identity.

[0043] Meanwhile, the computer vision recognition module continuously records the establishment and termination times of the association between teaching equipment and teachers, and generates equipment usage records based on equipment location change information and personnel identity information, thereby forming a complete teaching equipment usage trajectory, providing basic data support for subsequent cross-validation of course schedules and equipment responsibility traceability.

[0044] In this embodiment, the timetable data processing module is used to establish a data communication connection with the school's academic affairs system, and to collect, parse, store and manage the course data in the academic affairs system, providing basic data support for the identity verification of teaching equipment users and the traceability of equipment.

[0045] Specifically, the timetable data processing module connects with the school's academic affairs system through interface calls, database access, or data synchronization to obtain course arrangement information, teacher information, and classroom mapping relationships corresponding to the teaching scenario in real time or at regular intervals, and establishes a timetable database.

[0046] The course schedule information includes course name, teaching time, instructor information, course number, and classroom information, which is used to determine the course situation in the corresponding classroom within a specified time period. The instructor information includes instructor name, instructor ID, department, and facial feature information corresponding to the instructor's identity, which is used to match and verify the instructor's identity with the instructor identification results output by the computer vision recognition module. The classroom mapping relationship includes the correspondence between classroom number, classroom name, and surveillance camera identification information, which is used to establish the association between surveillance video data and the actual teaching scenario.

[0047] The timetable data processing module can also standardize and structure the acquired data, and dynamically update the timetable database based on information such as course adjustments, temporary class rescheduling, and teacher substitutions, thereby ensuring that the course data is consistent with the actual teaching situation and providing accurate data basis for the subsequent cross-validation module to verify teacher identity and generate teaching equipment usage records.

[0048] In this embodiment, the cross-validation module is used to verify the correlation between computer vision recognition results and timetable data in order to determine the actual users of the teaching equipment.

[0049] Specifically, once the computer vision recognition module detects the use of teaching equipment, it sends the corresponding teacher identity information, teaching equipment location information, and time information to the cross-validation module. The cross-validation module then queries the timetable database for course schedule information within the corresponding time period based on the time information and classroom location information of the teaching equipment, and obtains the information of the instructor for the course in that classroom.

[0050] Subsequently, the cross-validation module matches the teacher identity information output by the computer vision recognition module with the teaching teacher information in the course schedule. When they match, the teacher is determined to be the actual user of the teaching equipment, and a corresponding teaching equipment usage record is generated. When they do not match, the cross-validation module further retrieves course change information, substitute teaching information, and class rescheduling information for secondary verification to determine whether there are any course adjustments, substitute teaching, or temporary class changes.

[0051] If the second verification result can identify the actual teacher, the teaching equipment usage record will be updated according to the verification result, and the corresponding teacher will be identified as the actual user of the teaching equipment; if the actual user cannot be identified after the second verification, an abnormal verification record will be generated and sent to the management terminal for manual review.

[0052] When teaching equipment is lost, damaged, or there is a need to trace responsibility, the cross-validation module can retrieve the time and location information of the last time the target teaching equipment was identified, and perform correlation analysis by combining the course schedule information, teacher identity information, and historical usage records within the corresponding time period, thereby determining the last person to use the teaching equipment, providing a basis for tracing the source of teaching equipment and determining responsibility.

[0053] The equipment feature library module is used to store and manage the equipment feature information of various teaching equipment, and to perform feature matching and assisted identification of teaching equipment based on at least one of the equipment appearance features, identification features, RFID features and behavioral features to determine the identity information of teaching equipment; at the same time, it receives the equipment identification information sent by the RFID reading device and combines the computer vision recognition results to perform fusion verification of the identity of teaching equipment.

[0054] The traceability report module is used to receive the verification results output by the cross-validation module, and construct a traceability information chain for the equipment based on the teaching equipment usage records, teacher identity information, timetable verification results, and abnormal event information; generate the equipment usage trajectory based on the location data and user data of the teaching equipment at different time points, and determine the circulation process and responsibility of the teaching equipment in combination with the equipment traceability information chain, thereby generating a traceability report for the teaching equipment.

[0055] In this embodiment, the device feature information includes at least one of appearance features, identification features, RFID features, and behavioral features.

[0056] In this embodiment, the deep learning target detection model in the computer vision recognition module is the YOLO series target detection model, which realizes the target detection of teaching equipment by recognizing the appearance features of the equipment in the video image; S21. Preprocess the acquired video stream images and input them into the YOLO object detection network for feature extraction; S22. Multi-scale feature extraction of target regions in video images is performed through a backbone network to obtain candidate target feature information of teaching equipment. S23. Based on the candidate target feature information, predict the bounding box of the teaching equipment target to obtain the position parameters of the teaching equipment target. The bounding box prediction formula is: as well as: in, It is the Sigmoid activation function. The x-coordinate of the center point of the prediction box. This represents the lateral offset predicted by the network. This represents the vertical offset predicted by the network. The x-coordinate of the top-left corner of the current grid cell. The ordinate of the top-left corner of the current grid cell. To predict the box width, To predict the bounding box height, The width of the prior bounding box (Anchor). The height of the prior bounding box (Anchor) The width scaling parameter is obtained from the network prediction. Here, e is the height scaling parameter predicted by the network, and e is a natural constant, approximately equal to 2.71828. S24. Calculate the target confidence score based on the bounding box prediction results to determine whether the teaching equipment target exists. The target confidence score calculation formula is as follows: The formula for calculating the Intersection over Union (IoU) is: ,in, This represents the area of ​​the intersection region between the predicted bounding box and the ground truth bounding box. Let Pr(Object) be the area of ​​the union region of the predicted bounding box and the ground truth bounding box, and let loU(pred,truth) be the intersection-union ratio of the predicted bounding box and the ground truth bounding box. S25. Iteratively train the YOLO object detection model based on the loss function to improve the object detection accuracy of teaching equipment. The loss function adopted is CIoU Loss, and its expression is: The bounding box loss function is: ,in, For the total loss function, To locate the loss weights, Weight the target confidence level loss. For classification loss weights, For bounding box loss, For target confidence loss, For classifying losses, This is the distance between the center point of the predicted bounding box and the center point of the ground truth bounding box. This is the diagonal length of the smallest bounding rectangle between the predicted bounding box and the ground truth bounding box. This parameter ensures consistency in aspect ratio between the predicted bounding box and the ground truth bounding box. The aspect ratio weighting coefficient; S26. Output the target detection results of the teaching equipment to obtain the equipment category information, location information and corresponding time information of the teaching equipment.

[0057] In this embodiment, the human pose estimation algorithm is the OpenPose algorithm, which determines the position of the hand region by detecting key points on the human body and establishes a correspondence between the teaching equipment and the user. Its steps include: S31. Perform human target detection on the acquired video images and input the human images into the OpenPose human pose estimation network; S32. Based on the OpenPose human pose estimation network, human key points are detected, and confidence maps corresponding to each human key point are generated. The calculation formula for the key point confidence map is as follows: ,in, Indicates the first The first personal physical goal The confidence values ​​corresponding to each key point; p represents the coordinates of any pixel in the image; Indicates the first The first personal physical goal The actual coordinates of each key point; This represents the squared Euclidean distance between a pixel and its corresponding keypoint. The standard deviation of the Gaussian distribution is used to control the diffusion range of the influence area of ​​key points. Represents an exponential function; S33. Generate Part Affinity Fields (PAF) vector fields based on the connection relationships between key points of the human body to establish the structural association of human limbs. The calculation formula for the Part Affinity Fields is as follows: ,in, Indicates the body limb area Corresponding PAF supervision information; This represents a human limb region consisting of two key points; p represents a pixel within the limb region. This represents the predicted vector of the PAF vector field at the corresponding pixel. and These represent the coordinates of key points at both ends of the limb; Indicates key points Pointing to key points The unit direction vector; S34. Obtain the position of the key point of the human wrist based on the human key point detection results, and obtain the center position of the teaching equipment in combination with the target detection results of the teaching equipment. S35. Calculate the Euclidean distance between the center position of the teaching equipment and the key point position of the human wrist. The distance calculation formula is as follows: ,in, The coordinates of the center position of the teaching equipment Coordinates of key points on the human wrist; S36. When the Euclidean distance satisfies the following correlation condition: Then it is determined that there is a possession relationship between the corresponding personnel and the teaching equipment, whereby... Here, d represents the distance threshold, indicating the distance between the teaching device and the wrist key point. S37. Based on the holding relationship, establish the correspondence between teaching equipment and users, and output the equipment-person association result.

[0058] In this embodiment, the device feature library module uses an adaptive weight allocation multi-feature fusion strategy to identify teaching devices. The multi-feature fusion strategy includes: The feature weights are dynamically calculated based on the confidence levels corresponding to each device feature. The formula for calculating the feature weights is as follows: ,in, Indicates the first The weight value corresponding to each device feature is used to reflect the degree of contribution of that feature to the final recognition result; Indicates the first The confidence level of each device feature is used to characterize the reliability of the current feature recognition result; This represents the sum of confidence levels of all device features involved in the fusion, where Indicates the sequence numbers of all features involved in the fusion calculation; When the confidence level of the corresponding device feature is lower than the preset low confidence threshold, a reliability penalty is applied to the weight of the corresponding feature. The final identification result is obtained by fusing multiple device features using a weighted voting method. The fusion decision formula is as follows: ,in, This represents the final recognition result or comprehensive matching score after multi-feature fusion. This represents the weight value corresponding to the i-th device feature; Indicates the first The identification results, matching scores, or feature determination values ​​of each device feature; This indicates that the recognition results of all device features are weighted and summed according to their corresponding weights.

[0059] In this embodiment, the device feature library module is also used to reset the corresponding feature weight to zero and renormalize the weights of the remaining features when a single device feature fails; and to output unrecognizable information and trigger manual processing when all device features fail. The device feature library module also judges the recognition result based on a preset decision threshold. When the final recognition result is greater than the acceptance threshold, the recognition pass result is output. When the final recognition result is lower than the rejection threshold, the recognition failure result is output. When the final recognition result is between the acceptance threshold and the rejection threshold, the multi-feature fusion review flow is triggered.

[0060] In this embodiment, the computer vision recognition module is also used to adaptively optimize video images for low light, target occlusion, and multiple people in the same frame in teaching scenarios, so as to improve the accuracy of target recognition and personnel identification of teaching equipment.

[0061] The computer vision recognition module performs scene optimization processing for low-light conditions, target occlusion, and multiple people in the same frame in teaching scenarios. In low-light scenes, the computer vision recognition module enhances the brightness of the input image using the CLAHE adaptive histogram equalization algorithm, and the enhancement formula is as follows: , among which, among which, This represents the image after enhancement processing; I represents the original input image captured by the camera. This represents the contrast-limited adaptive histogram equalization processing function.

[0062] Next, the image reflection component is enhanced based on the Retinex enhancement algorithm, and the enhancement formula is as follows: ,in, Indicates the image in coordinates Enhancement results at the location; This represents the brightness information of the input image at the corresponding location; This represents the illumination estimation component generated by Gaussian filtering; and These represent the horizontal and vertical coordinates in the image, respectively. Simultaneously, infrared and visible light images are fused to improve the device's detection capabilities under low-light conditions; the fusion formula is as follows: ,in, This represents the merged image; Represents a visible light image; Represents an infrared image; This represents the fusion weighting coefficient, which ranges from 0 to... This represents the weight corresponding to the infrared image.

[0063] In this embodiment, the computer vision recognition module generates an occlusion region mask through an occlusion detection model in a target occlusion scenario to identify the visible area of ​​the target. The occlusion detection formula is as follows: ,in, Indicates the masking area; Indicates the input image; This represents the occlusion detection model. Based on the occlusion region mask, features of the visible target region are extracted. The feature extraction formula is as follows: ,in, Indicates the characteristics of the visible area of ​​the target; Represents the feature extraction function; Indicates the input image; Indicates the masking area; Indicates the visible area mask; This represents element-wise multiplication. Simultaneously, the computer vision recognition module calculates the target matching score using a partial feature matching method, with the following scoring formula: ,in, This indicates the overall matching score of the target. Indicates the first The weights corresponding to each feature; The first term of the target to be identified One feature; This represents the first target in the device feature library. One feature; This represents the function for calculating feature similarity. In this embodiment, the computer vision recognition module, in a multi-person scene, sorts the detected faces according to their face area and prioritizes recognizing larger face targets. Simultaneously, it determines the person holding the device based on the distance between the teaching equipment location and the face location. The selection formula is as follows: ,in, This represents the final determined personnel holding the equipment; argmin represents the target corresponding to the minimum value. Indicates the coordinates of the center position of the face; Indicates the coordinates of the center position of the teaching equipment; The Euclidean distance between the two is represented, and the DeepSORT multi-object tracking algorithm is used to maintain the consistency of personnel IDs; the computer vision recognition module also matches target features based on cosine similarity, and the matching formula is: ,in, Indicates the similarity between feature vectors; This represents the feature vector of the target to be identified; This represents a reference feature vector in the feature library; Represents the dot product of two vectors; Representing vectors The modulus length; Representing vectors The length of the module.

[0064] Different recognition thresholds are used for different teaching scenarios, with a recognition threshold of 0.6 for normal scenarios, 0.55 for low-light scenarios, and 0.65 for occluded scenarios.

[0065] In this embodiment, the device feature library module uses an adaptive weight allocation multi-feature fusion strategy to identify teaching equipment, including: The feature weights are dynamically calculated based on the confidence levels corresponding to each device feature. The formula for calculating the feature weights is as follows: ,in, Indicates the first The weight values ​​corresponding to each device feature; Indicates the first Confidence level of each device feature; The first to participate in the integration The confidence level of each device feature is calculated; by normalizing the confidence level of each device feature, the contribution weight of different features in the final recognition result is dynamically allocated.

[0066] When the confidence level corresponding to the i-th device feature When the confidence level is below the preset low confidence threshold Tlow, a reliability penalty correction is applied to the corresponding feature weights so that the corrected feature weights satisfy the following: ,in, This represents the feature weights after reliability penalty; This represents the feature weights before penalty; 0.5 represents the weight decay coefficient; and when the confidence level of the corresponding feature is... Below the preset low confidence threshold When this happens, the weights of the features are attenuated to reduce the impact of low-reliability features on the final recognition result.

[0067] The fusion decision is based on the weights corresponding to the features of each device and the recognition results. A weighted voting method is used to obtain the final recognition result. The fusion decision formula is as follows: ,in, This represents the final recognition result after multi-feature fusion; Indicates the first The weights corresponding to each device feature; Indicates the first The recognition result or matching score corresponding to each device feature.

[0068]

[0069] Example 1: Tracing the source of a lost UHF wireless microphone A middle school has equipped several classrooms with UHF wireless microphones for teachers to use during lessons. Each microphone has a unique device number. One day, the administrator discovered that the UHF wireless microphone with the serial number MIC-003 was missing and needed to determine the last person to use the device.

[0070] After the administrator initiates a device traceability request through the teaching equipment traceability system of this invention, the traceability report module first retrieves the historical usage records corresponding to MIC-003. The computer vision recognition module retrieves the video data collected by the surveillance camera and finds that the device was last identified at 14:25 on March 10, 2026, located in classroom 3. Through device detection, human posture estimation, and facial recognition, it is determined that the person holding the device at that time was Teacher Zhang (employee number T2021003).

[0071] Subsequently, the cross-validation module queried the timetable database based on the device's corresponding time information and classroom location information to obtain the course schedule information for classroom 3 from 14:00 to 15:40 on March 10, 2026. Comparison revealed that the instructor for the courses during this time period was Teacher Zhang, and the timetable information matched the visual recognition results. Therefore, the cross-validation module confirmed that Teacher Zhang was the last user of MIC-003 and generated the corresponding teaching equipment usage record.

[0072] Finally, the traceability report module generates a traceability report for teaching equipment based on equipment usage records, teacher identity information, course schedule verification results, and surveillance video evidence. The report includes equipment number, last usage time, usage location, teacher identity information, surveillance screenshots, course schedule records, and verification results, thus providing a basis for equipment management and accountability.

[0073] Example 2: Tracing the Borrowing of Teaching Page Turners A university's public teaching building has laser pointers installed at the podiums of each classroom for teachers to use during lectures. One day, the administrator discovered that a laser pointer was missing from classroom 205 and needed to trace its usage.

[0074] After receiving the device query request, the system analyzes the surveillance video of classroom 205 using a computer vision recognition module and confirms the device's identity based on the page turner's feature information in the device feature database. The system identifies the page turner as having been last detected at 10:15 AM on March 12, 2026, and determines that the last person to possess the device was Teacher Li.

[0075] Subsequently, the cross-validation module queried the timetable database based on the device's corresponding time and classroom information, and obtained the course schedule information for classroom 205 within the corresponding time period. Verification confirmed that the instructor for the 3rd and 4th periods on March 12, 2026, was indeed Ms. Li; therefore, the system confirmed that Ms. Li was the last user of the page turner.

[0076] After confirming the results, the traceability report module automatically generates the equipment usage trajectory and sends the last usage time, location, user, and course verification results to the equipment management terminal, so that managers can contact the corresponding teacher in a timely manner to verify the equipment and improve equipment management efficiency.

[0077] Example 3: Traceability of Equipment Liability in Case of Course Abnormalities After a teaching equipment loss incident, the system used a computer vision recognition module to identify Mr. Wang as the last person to possess the equipment. However, a cross-validation module querying the timetable database revealed that Mr. Zhao was the instructor for the corresponding class in that classroom during that time period, indicating a discrepancy between the visual recognition result and the course schedule information.

[0078] At this point, the cross-validation module automatically identified the event as an abnormal validation event and further retrieved course change records, substitute teaching records, and class rescheduling records for secondary verification. The verification revealed that Teacher Zhao was unable to teach due to a last-minute leave request, and Teacher Wang had filled in for her, but the relevant course adjustment information had not yet been synchronized to the course schedule database.

[0079] Based on the above verification results, the cross-validation module automatically updates the teaching equipment usage record, confirming Teacher Wang as the actual user of the equipment, and records the reason for the course anomaly. Simultaneously, the system generates an anomaly verification record and sends the verification results to the management terminal, prompting administrators to update the course schedule data accordingly.

[0080] Ultimately, the traceability report module generates a complete traceability report for abnormal events based on visual recognition results, course schedule data, substitute teaching records, and manual verification results, thereby ensuring that the responsibility for teaching equipment can still be accurately traced in special teaching scenarios such as course adjustments and temporary substitute teaching.

[0081] This invention is not limited to the embodiments described above. Anyone should understand that structural changes made under the guidance of this invention, and any technical solutions that are the same as or similar to this invention, fall within the protection scope of this invention. Technical aspects, shapes, and structures not described in detail in this invention are all publicly known technologies.

Claims

1. A teaching equipment traceability system based on cross-validation of computer vision and timetables, characterized in that, include: The computer vision recognition module is used to collect video data of teaching scenarios and identify teaching equipment and information about its users. The timetable data processing module is used to obtain course schedule information, teacher information, and classroom information; The equipment feature library module is used to store feature information of teaching equipment and to identify the teaching equipment. The cross-validation module is used to perform correlation verification based on the teaching equipment information, user information, and course schedule information to determine the actual user of the teaching equipment and generate equipment usage records. The traceability report module is used to generate a teaching equipment usage trajectory based on the equipment usage records and output a teaching equipment traceability report.

2. The teaching equipment traceability system based on cross-validation of computer vision and timetables as described in claim 1, characterized in that, The device feature information includes at least one of appearance features, identification features, RFID features, and behavioral features.

3. The teaching equipment traceability system based on cross-validation of computer vision and timetables as described in claim 1, characterized in that, The deep learning target detection model in the computer vision recognition module is the YOLO series target detection model, which realizes the target detection of teaching equipment by recognizing the appearance features of the equipment in the video image; S21. Preprocess the acquired video stream images and input them into the YOLO object detection network for feature extraction; S22. Multi-scale feature extraction of target regions in video images is performed through a backbone network to obtain candidate target feature information of teaching equipment. S23. Based on the candidate target feature information, predict the bounding box of the teaching equipment target to obtain the position parameters of the teaching equipment target. The bounding box prediction formula is: as well as: in, It is the Sigmoid activation function. The x-coordinate of the center point of the prediction box. This represents the lateral offset predicted by the network. This represents the vertical offset predicted by the network. The x-coordinate of the top-left corner of the current grid cell. The ordinate of the top-left corner of the current grid cell. To predict the box width, To predict the bounding box height, The width of the prior bounding box (Anchor). The height of the prior bounding box (Anchor) The width scaling parameter is obtained from the network prediction. The height scaling parameter is obtained from the network prediction. S24. Calculate the target confidence score based on the bounding box prediction results to determine whether the teaching equipment target exists. The target confidence score calculation formula is as follows: The formula for calculating the Intersection over Union (IoU) is: ,in, This represents the area of ​​the intersection region between the predicted bounding box and the ground truth bounding box. Let Pr(Object) be the area of ​​the union region of the predicted bounding box and the ground truth bounding box, and let loU(pred,truth) be the intersection-union ratio of the predicted bounding box and the ground truth bounding box. S25. Iteratively train the YOLO object detection model based on the loss function to improve the object detection accuracy of teaching equipment. The loss function adopted is CIoU Loss, and its expression is: The bounding box loss function is: ,in, For the total loss function, To locate the loss weights, Weight the target confidence level loss. For classification loss weights, For bounding box loss, For target confidence loss, For classifying losses, This is the distance between the center point of the predicted bounding box and the center point of the ground truth bounding box. This is the diagonal length of the smallest bounding rectangle between the predicted bounding box and the ground truth bounding box. This parameter ensures consistency in aspect ratio between the predicted bounding box and the ground truth bounding box. The aspect ratio weighting coefficient; S26. Output the target detection results of the teaching equipment to obtain the equipment category information, location information and corresponding time information of the teaching equipment.

4. The teaching equipment traceability system based on cross-validation of computer vision and timetable as described in claim 1, characterized in that, The human pose estimation algorithm is the OpenPose algorithm, which determines the position of the hand region by detecting key points on the human body and establishes a correspondence between the teaching equipment and the user. Its steps include: S31. Perform human target detection on the acquired video images and input the human images into the OpenPose human pose estimation network; S32. Based on the OpenPose human pose estimation network, human key points are detected, and confidence maps corresponding to each human key point are generated. The calculation formula for the key point confidence map is as follows: ,in, Indicates the first The first personal physical goal The confidence values ​​corresponding to each key point; p represents the coordinates of any pixel in the image; Indicates the first The first personal physical goal The actual coordinates of each key point; This represents the squared Euclidean distance between a pixel and its corresponding keypoint. The standard deviation of the Gaussian distribution is used to control the diffusion range of the influence area of ​​key points. Represents an exponential function; S33. Generate Part Affinity Fields (PAF) vector fields based on the connection relationships between key points of the human body to establish the structural association of human limbs. The calculation formula for the Part Affinity Fields is as follows: ,in, Indicates the body limb area Corresponding PAF supervision information; This represents a human limb region consisting of two key points; p represents a pixel within the limb region. This represents the predicted vector of the PAF vector field at the corresponding pixel. and These represent the coordinates of key points at both ends of the limb; Indicates key points Pointing to key points The unit direction vector; S34. Obtain the position of the key point of the human wrist based on the human key point detection results, and obtain the center position of the teaching equipment in combination with the target detection results of the teaching equipment. S35. Calculate the Euclidean distance between the center position of the teaching equipment and the key point position of the human wrist. The distance calculation formula is as follows: ,in, The coordinates of the center position of the teaching equipment Coordinates of key points on the human wrist; S36. When the Euclidean distance satisfies the following correlation condition: Then it is determined that there is a possession relationship between the corresponding personnel and the teaching equipment, whereby... Here, d represents the distance threshold, indicating the distance between the teaching device and the wrist key point. S37. Based on the holding relationship, establish the correspondence between teaching equipment and users, and output the equipment-person association result.

5. The teaching equipment traceability system based on cross-validation of computer vision and timetables as described in claim 1, characterized in that, The device feature library module uses an adaptive weight allocation multi-feature fusion strategy to identify teaching devices. The multi-feature fusion strategy includes: The feature weights are dynamically calculated based on the confidence levels corresponding to each device feature. The formula for calculating the feature weights is as follows: ,in, Indicates the first The weight value corresponding to each device feature is used to reflect the degree of contribution of that feature to the final recognition result; Indicates the first The confidence level of each device feature is used to characterize the reliability of the current feature recognition result; This represents the sum of confidence levels of all device features involved in the fusion, where Indicates the sequence numbers of all features involved in the fusion calculation; When the confidence level of the corresponding device feature is lower than the preset low confidence threshold, a reliability penalty is applied to the weight of the corresponding feature. The final identification result is obtained by fusing multiple device features using a weighted voting method. The fusion decision formula is as follows: ,in, This represents the final recognition result or comprehensive matching score after multi-feature fusion. This represents the weight value corresponding to the i-th device feature; Indicates the first The identification results, matching scores, or feature determination values ​​of each device feature; This indicates that the recognition results of all device features are weighted and summed according to their corresponding weights.

6. The teaching equipment traceability system based on cross-validation of computer vision and timetable as described in claim 1, characterized in that, The device feature library module is also used to reset the corresponding feature weight to zero and renormalize the weights of the remaining features when a single device feature fails. When all device features fail, an unidentifiable message is output and a manual processing procedure is triggered. The device feature library module also judges the recognition result based on a preset decision threshold. When the final recognition result is greater than the acceptance threshold, the recognition pass result is output. When the final recognition result is lower than the rejection threshold, the recognition failure result is output. When the final recognition result is between the acceptance threshold and the rejection threshold, the multi-feature fusion review flow is triggered.

7. The teaching equipment traceability system based on cross-validation of computer vision and timetable as described in claim 1, characterized in that, The computer vision recognition module is also used to adaptively optimize video images in low light, target occlusion, and multiple people in the same frame in teaching scenarios, so as to improve the accuracy of target recognition and personnel identification of teaching equipment. The computer vision recognition module performs scene optimization processing for low-light conditions, target occlusion, and multiple people in the same frame in teaching scenarios. In low-light scenes, the computer vision recognition module enhances the brightness of the input image using the CLAHE adaptive histogram equalization algorithm, and the enhancement formula is as follows: , Among them, among them, This represents the image after enhancement processing; I represents the original input image captured by the camera. This represents a contrast-limited adaptive histogram equalization function. Next, the image reflection component is enhanced based on the Retinex enhancement algorithm, and the enhancement formula is as follows: ,in, Indicates the image in coordinates Enhancement results at the location; This represents the brightness information of the input image at the corresponding location; This represents the illumination estimation component generated by Gaussian filtering; and These represent the horizontal and vertical coordinates in the image, respectively. Simultaneously, infrared and visible light images are fused to improve the device's detection capabilities under low-light conditions; the fusion formula is as follows: ,in, This represents the merged image; Represents a visible light image; Represents an infrared image; This represents the fusion weighting coefficient, which ranges from 0 to... This represents the weight corresponding to the infrared image.

8. The teaching equipment traceability system based on cross-validation of computer vision and timetable as described in claim 7, characterized in that, In a target occlusion scenario, the computer vision recognition module generates an occlusion region mask using an occlusion detection model to identify the visible area of ​​the target. The occlusion detection formula is as follows: ,in, Indicates the masking area; Indicates the input image; This represents an occlusion detection model, which extracts features of the visible area of ​​the target based on an occlusion region mask. The feature extraction formula is as follows: ,in, Indicates the characteristics of the visible area of ​​the target; Represents the feature extraction function; Indicates the input image; Indicates the masking area; Indicates the visible area mask; This represents element-wise multiplication. Simultaneously, the computer vision recognition module calculates the target matching score using a partial feature matching method, with the following scoring formula: ,in, This indicates the overall matching score of the target. Indicates the first The weights corresponding to each feature; The first term of the target to be identified One feature; This represents the first target in the device feature library. One feature; This represents the function for calculating feature similarity.

9. A teaching equipment traceability system based on cross-validation of computer vision and timetables as described in claim 7, characterized in that, In scenarios with multiple people in the same frame, the computer vision recognition module sorts the detected faces according to their area, prioritizing the recognition of larger faces. Simultaneously, it determines the person holding the device based on the distance between the teaching equipment and the face, using the following formula: ,in, This represents the final determined personnel holding the equipment; argmin represents the target corresponding to the minimum value. Indicates the coordinates of the center position of the face; Indicates the coordinates of the center position of the teaching equipment; The Euclidean distance between the two is represented, and the DeepSORT multi-object tracking algorithm is used to maintain the consistency of personnel IDs; the computer vision recognition module also matches target features based on cosine similarity, and the matching formula is: ,in, Indicates the similarity between feature vectors; This represents the feature vector of the target to be identified; This represents a reference feature vector in the feature library; Represents the dot product of two vectors; Representing vectors The modulus length; Representing vectors The modulus length; Different recognition thresholds are used for different teaching scenarios, with a recognition threshold of 0.6 for normal scenarios, 0.55 for low-light scenarios, and 0.65 for occluded scenarios.

10. A teaching equipment traceability system based on cross-validation of computer vision and timetables as described in claim 1, characterized in that, The device feature library module employs an adaptive weight allocation multi-feature fusion strategy to identify teaching equipment, including: The feature weights are dynamically calculated based on the confidence levels corresponding to each device feature. The formula for calculating the feature weights is as follows: ,in, Indicates the first The weight values ​​corresponding to each device feature; Indicates the first Confidence level of each device feature; The first to participate in the integration Confidence level of each device feature; When the confidence level corresponding to the i-th device feature When the confidence level is below the preset low confidence threshold Tlow, a reliability penalty correction is applied to the corresponding feature weights so that the corrected feature weights satisfy the following: ,in, This represents the feature weights after reliability penalty; 0.5 represents the feature weights before penalty; 0.5 represents the weight decay coefficient. The fusion decision is based on the weights corresponding to the features of each device and the recognition results. A weighted voting method is used to obtain the final recognition result. The fusion decision formula is as follows: ,in, This represents the final recognition result after multi-feature fusion; Indicates the first The weights corresponding to each device feature; Indicates the first The recognition result or matching score corresponding to each device feature.

Images