An Improved Customer Flow Statistics Method for 4S Stores Based on Pedestrian Re-identification

Abstract

This invention relates to the field of computer vision technology, specifically to an improved 4S store customer flow statistics method based on pedestrian re-identification. The method includes the following steps: real-time acquisition of video data from the 4S store entrance using an image acquisition device; standardization of the video data based on video processing input requirements to generate standardized video processing results. In this invention, real-time acquisition and standardization of video data ensures image quality and data processing continuity under high customer flow conditions. Adjustments to the size and position of target detection boxes improve the accuracy of target recognition and tracking. The method analyzes the movement paths of targets and adds marker information to accurately track the movements of multiple customers. Comparison with feature vectors stored in a database effectively distinguishes duplicate targets and performs customer flow statistics. Combined with entry and exit behavior analysis based on store layout parameters, this helps the store adjust its human resource allocation, improving operational efficiency and customer satisfaction.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to an improved 4S store customer flow statistics method based on pedestrian re-identification. Background Technology

[0002] The field of computer vision technology utilizes artificial intelligence and deep learning to analyze image data captured by cameras and sensors, enabling machines to understand visual information. Through methods, it parses image data to identify and classify objects, scenes, and activities, performing various tasks such as image processing, object detection, image classification, and image reconstruction without human intervention. It is applied to a variety of practical scenarios, including autonomous driving, industrial automation, medical imaging analysis, security monitoring, and human-computer interaction. By improving and optimizing the accuracy and processing speed of the methods, the system can operate stably in a variety of complex environments and conditions.

[0003] Among them, the 4S store customer flow statistics method focuses on customer flow statistics for 4S car showrooms and service stores. The system uses target detection, multi-target tracking, and pedestrian re-identification technologies in computer vision to accurately calculate the number of customers entering and leaving the store. By distinguishing and tracking each customer, it maintains a high accuracy rate even when there are a large number of customers and frequent entry and exit. It can be applied to business environments that require accurate customer flow data to optimize store operations and improve customer service experience, manage customer traffic, rationally allocate personnel and resources, and improve operational efficiency and customer satisfaction.

[0004] Traditional 4S store customer flow statistics technology has shortcomings in the field of commercial customer flow statistics. It relies on simple image processing technology, which makes it susceptible to the effects of light and occlusion in high-traffic and complex environments. When processing large amounts of real-time data, it requires high computing resources, which limits its application in resource-constrained environments. It lacks effective customer behavior tracking and analysis capabilities, making it difficult to provide merchants with in-depth customer behavior insights. The inability to distinguish between new and old customers makes it difficult for merchants to effectively implement targeted marketing strategies, affecting the application effect of customer flow data by commercial users, limiting the flexibility and responsiveness of business operations, and reducing customer satisfaction and the economic benefits of merchants. Summary of the Invention

[0005] To address the technical problems inherent in existing technologies, such as high labor costs, insufficient equipment funding, and limited installation space, which hinder their ability to meet the growing demands of modern 4S stores, this invention provides an improved 4S store customer flow statistics method based on pedestrian re-identification. The technical solution is as follows:

[0006] On the one hand, an improved 4S store customer flow statistics method based on pedestrian re-identification is provided. This method is implemented by an SSS device and includes:

[0007] S1: Based on image acquisition equipment, real-time video data of the 4S store entrance is collected. Combined with the input requirements of video processing, the video data is standardized to generate standardized video processing results.

[0008] S2: Based on the video standardization processing results, by configuring the learning speed to match the dynamic changes in the video, objects in the video data are detected, and the size and position of the recognition box are adjusted to reduce the overlap between multiple boxes, thereby obtaining target detection box information;

[0009] S3: Based on the target detection box information, by analyzing the position parameters of the identified target, the position change of the target between consecutive frames is calculated, the motion path of the target is identified and marker information is added to generate motion path tracking information;

[0010] S4: Based on the mobile path tracking information, feature vectors of multiple targets are extracted and compared with feature vectors stored in the database to calculate the similarity score between feature data and obtain the re-identification and classification result;

[0011] S5: Based on the re-identification and classification results, combined with the store layout parameter settings, enter the detection buffer, analyze the entry and exit behavior of multiple targets in real time, update and record the entry and exit status of multiple targets, and generate the area entry detection results;

[0012] S6: Based on the area entry detection results, count the pedestrians entering and exiting the store entrance, record the characteristics and identification information of multiple pedestrians, assess the manpower configuration required for the operation of the target store, and generate store customer flow statistics.

[0013] As a further aspect of the present invention, the video standardization processing result includes image resolution adjustment information, frame rate adjustment records, and timestamp records; the target detection box information includes positioning and recognition box size parameters, recognition box position information, and recognition box positioning position information; the movement path tracking information includes target motion speed data, motion direction information, and time series of motion trajectory; the re-identification and classification result includes target feature similarity score, a list of confirmed targets, and feature library matching information of targets to be re-identified; the area entry detection result includes target entry and exit status records, behavior analysis of targets within the buffer zone, and spatial position change records of targets; and the store customer flow statistics include the number of counted pedestrians, pedestrian feature and identification dataset, and customer flow change trend data.

[0014] As a further aspect of the present invention, the steps of acquiring video data from the entrance of a 4S store in real time using an image acquisition device, and standardizing the video data to generate a standardized video processing result, based on the input requirements of video processing, are as follows:

[0015] S101: Based on the image acquisition device, by adjusting the focal length and viewing angle of the camera, the acquisition range is optimized, video data of the 4S store entrance is acquired in real time, and the time information of the acquisition device is calibrated to optimize the consistency between video data and actual time, and video acquisition results are generated.

[0016] S102: Based on the video acquisition results, and according to the input requirements of video processing, adjust the resolution of multiple frames in the video, optimize image clarity, and generate a resolution-adjusted video stream;

[0017] S103: Adjust the video stream based on the resolution, adjust the frame rate of the video stream, set the frame rate to match the dynamic scene changes of the video surveillance, and generate a video standardization processing result.

[0018] As a further aspect of the present invention, based on the video standardization processing result, by configuring the learning speed to match the dynamic changes in the video, objects in the video data are detected, and the size and position of the recognition boxes are adjusted to reduce the overlap between multiple boxes to obtain the target detection box information, the specific steps are as follows:

[0019] S201: Based on the video standardization processing results, adjust the learning rate of the target detection model, match the dynamic changes in the video data, optimize the model's response speed to various scenarios, and generate learning rate adjustment results;

[0020] S202: Based on the learning rate adjustment result, detect moving objects in the video frame, record the position and movement information of multiple targets, and generate target position data;

[0021] S203: Based on the target location data, reduce the overlap between boxes by adjusting the size and position of the recognition box, optimize the recognition accuracy and target tracking ability, and generate target detection box information.

[0022] As a further aspect of the present invention, the specific formula for adjusting the learning rate of the target detection model is as follows:

[0023]

[0024] in, The learning rate adjustment function is used to calculate the current iteration number. The learning rate function incorporates cosine annealing to optimize dynamic adjustment of the learning rate, ensuring a larger learning rate in the early stages of training for rapid convergence, and gradually decreasing it towards the end of training to refine the adjustment of model parameters. This represents the current iteration number, indicating a point in time during the learning process, and is used to adjust the learning rate in real time. The total number of iterations is defined as the preset total number of iteration cycles during the entire training process, used for normalization. The ratio is adjusted to ensure a reasonable distribution of learning rate adjustments throughout the training cycle. This is the minimum learning rate, ensuring that the learning rate does not drop to an extremely low level that is ineffective for model training during the reduction process, thereby maintaining the effectiveness and stability of the method throughout the training process. Pi is a constant introduced in the calculation to ensure that the adjustment of the learning rate has a smooth, periodic transition, adapting to different training phases.

[0025] As a further aspect of the present invention, based on the target detection box information, the steps of analyzing the position parameters of the identified target, calculating the position change of the target between consecutive frames, identifying the target's motion path and adding marker information to generate motion path tracking information are as follows:

[0026] S301: Based on the target detection box information, analyze and identify the position coordinates of multiple targets, compare the coordinate data between consecutive frames, analyze the movement direction and speed of the targets in the video, and generate position change analysis results;

[0027] S302: Based on the position change analysis results, track the movement paths of multiple targets in real time, record the movement trajectory and duration of the targets in the video sequence, and generate trajectory recording results;

[0028] S303: Based on the trajectory recording results, identification markers are added to multiple targets by associating them with the target's motion trajectory and timestamp, generating movement path tracking information.

[0029] As a further aspect of the present invention, based on the mobile path tracking information, the specific steps of extracting feature vectors of multiple targets and comparing them with feature vectors already stored in the database to calculate the similarity score between feature data and obtain the re-identification and classification result are as follows:

[0030] S401: Based on the movement path tracking information, identify the feature vectors of multiple targets according to the image information, including the target size, geometric shape, and color information, and generate feature vector extraction results;

[0031] S402: Based on the feature vector extraction results, the similarity score between the two sets of vectors is calculated by comparing them with the targets identified in the database, and feature matching results are generated;

[0032] S403: Based on the feature matching results, by judging the matching status between the target and known targets in the database, the matched targets are classified and their identity labels are updated, the re-identification status of the targets is identified, and re-identification classification results are generated.

[0033] As a further aspect of the present invention, based on the re-identification and classification results, combined with the store layout parameter settings for entering the detection buffer, the steps of analyzing the entry and exit behaviors of multiple targets in real time, updating and recording the entry and exit status of multiple targets, and generating the area entry detection results are as follows:

[0034] S501: Based on the re-identification and classification results, analyze the layout of the target 4S store and configure the boundary parameters of the entrance buffer zone, including the spatial layout of the store entrance and the direction of pedestrian flow, and generate the buffer zone parameter configuration.

[0035] S502: Based on the buffer parameter configuration, the video data is analyzed in real time, the location data is used to determine and identify targets that cross the buffer, the entry and exit status of the targets is recorded, and a real-time entry and exit status record is generated.

[0036] S503: Based on the real-time entry and exit status record, update the target entry and exit data in the database in real time, including the identification code of the target and entry and exit time information, and generate the area entry detection result.

[0037] As a further aspect of the present invention, based on the area entry detection results, the steps of counting pedestrians entering and exiting the store entrance, recording the characteristics and identification information of multiple pedestrians, assessing the manpower allocation required for the operation of the target store, and generating store customer flow statistics are as follows:

[0038] S601: Based on the area entry detection results, calculate the pedestrian flow at the store entrance, including the number of people entering and leaving the store, and generate a pedestrian flow record;

[0039] S602: Based on the pedestrian flow record, record the features and identification information of multiple targets, including appearance features, timestamps, and entry and exit status, to provide data support for pedestrian re-identification and generate pedestrian feature analysis data;

[0040] S603: Based on the pedestrian feature analysis data, by performing time series analysis on the store's customer traffic data, calculate the changing trend of customer traffic, assess the manpower allocation required for the target store, and generate store customer traffic statistics.

[0041] As a further aspect of the present invention, the specific formula for calculating the changing trend of passenger flow is as follows:

[0042]

[0043] in, Represents a point in time The trend estimate is used to predict passenger flow at next point in time. Represents a point in time It provides the most accurate and current passenger flow data, reflecting actual passenger volume. Represents a point in time Actual passenger flow, provided slightly earlier Passenger flow information Represents a point in time The actual passenger flow helps in analyzing earlier changes in passenger flow. Represents a point in time The actual passenger flow is used as a reference for long-term trend analysis. yes The weighting coefficients determine the importance of the most recent data in the overall trend analysis. yes The weighting coefficients are used to adjust the contribution of the penultimate period data to the prediction model. yes The weighting coefficients affect the role of the penultimate period data in the overall analysis. yes The weighting coefficients are used to determine the influence of the earliest data in trend prediction.

[0044] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0045] By acquiring video data in real time and performing standardized processing, image quality and data processing continuity are ensured under high customer traffic conditions. Combined with adjustments to the size and position of the target detection box, the accuracy of target recognition and tracking is improved. The movement path of the target is analyzed and labeled, accurately tracking the movements of multiple customers. By comparing with the feature vectors stored in the database, duplicate targets are effectively distinguished and customer flow statistics are performed. Combined with the entry and exit behavior analysis implemented with store layout parameters, the store can adjust its human resource allocation, improve operational efficiency and customer satisfaction. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of the improved 4S store customer flow statistics method based on pedestrian re-identification provided in the embodiments of the present invention;

[0048] Figure 2 This is a detailed schematic diagram of S1 of the present invention;

[0049] Figure 3 This is a detailed schematic diagram of S2 of the present invention;

[0050] Figure 4 This is a detailed schematic diagram of S3 of the present invention;

[0051] Figure 5 This is a detailed schematic diagram of S4 of the present invention;

[0052] Figure 6 This is a detailed schematic diagram of S5 of the present invention;

[0053] Figure 7 This is a detailed schematic diagram of S6 of the present invention. Detailed Implementation

[0054] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0055] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0056] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0057] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0058] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0059] This invention provides an improved 4S store customer flow statistics method based on pedestrian re-identification, such as... Figure 1 The flowchart shown is for a 4S store customer flow statistics method improved based on pedestrian re-identification. The processing flow of this method may include the following steps:

[0060] S1: Based on image acquisition equipment, real-time video data of the 4S store entrance is collected. Combined with the input requirements of video processing, the video data is standardized to generate standardized video processing results.

[0061] S2: Based on the video standardization processing results, by configuring the learning speed to match the dynamic changes in the video, objects in the video data are detected, and the size and position of the recognition box are adjusted to reduce the overlap between multiple boxes, thereby obtaining target detection box information;

[0062] S3: Based on the target detection box information, by analyzing the position parameters of the identified target, the position change of the target between consecutive frames is calculated, the motion path of the target is identified and marker information is added to generate motion path tracking information;

[0063] S4: Based on the mobile path tracking information, feature vectors of multiple targets are extracted and compared with feature vectors stored in the database to calculate the similarity score between feature data and obtain the re-identification and classification result;

[0064] S5: Based on the re-identification and classification results, combined with the store layout parameter settings, enter the detection buffer, analyze the entry and exit behavior of multiple targets in real time, update and record the entry and exit status of multiple targets, and generate the area entry detection results;

[0065] S6: Based on the area entry detection results, count the pedestrians entering and exiting the store entrance, record the characteristics and identification information of multiple pedestrians, assess the manpower configuration required for the operation of the target store, and generate store customer flow statistics.

[0066] The video standardization processing results include image resolution adjustment information, frame rate adjustment records, and timestamp records. The target detection box information includes the location recognition box size parameters, recognition box position information, and recognition box location information. The movement path tracking information includes the target's motion speed data, motion direction information, and time series of motion trajectory. The re-identification and classification results include the target's feature similarity score, a list of confirmed targets, and feature library matching information of targets to be re-identified. The region entry detection results include target entry and exit status records, behavior analysis of targets within the buffer zone, and spatial position change records of targets. The store customer flow statistics include the number of counted pedestrians, pedestrian feature and identification datasets, and customer flow change trend data.

[0067] Please see Figure 2 Based on image acquisition equipment, real-time video data is collected at the entrance of the 4S store. Combined with the input requirements for video processing, the video data is standardized to generate standardized video processing results. The specific steps are as follows:

[0068] S101: Based on the image acquisition device, by adjusting the focal length and viewing angle of the camera, the acquisition range is optimized, video data of the 4S store entrance is acquired in real time, and the time information of the acquisition device is calibrated to optimize the consistency between video data and actual time, and video acquisition results are generated.

[0069] In sub-step S101, based on the image acquisition device, the operating parameters of the device are adjusted to improve data quality. By adjusting the focal length of the camera, it is ensured that vehicles and pedestrians at the entrance of the 4S store are clearly captured. The focal length adjustment depends on the result of calculating the average distance between the camera and the target object. Then, the viewing angle is adjusted to cover a wider acquisition range. The camera viewing angle adjustment takes into account the entrance width and traffic flow to determine the optimal shooting angle. The time information of the acquisition device is calibrated to ensure the consistency between the video data and the actual time. The calibration process includes synchronizing the internal clock of the device with the network time service to ensure that each frame of video can be correctly timestamped. The parameters are executed by the image acquisition control software and fine-tuned according to the feedback of the real-time monitored images to ensure that the acquired video data can reflect the dynamic situation at the entrance of the 4S store in real time.

[0070] S102: Based on the video acquisition results, and according to the input requirements of video processing, adjust the resolution of multiple frames in the video, optimize image clarity, and generate a resolution-adjusted video stream;

[0071] In sub-step S102, the video acquisition results are processed to match the subsequent video analysis requirements. The resolution of multiple frames of images is adjusted using bicubic interpolation to increase image resolution while preserving image details. During the processing, the method reads the image data of each frame, calculates the target resolution to be adjusted, resamples the original image pixels, and interpolates each pixel using the weighted values ​​of the surrounding pixels. The resulting high-resolution image stream is more suitable for human recognition. The adjusted video stream is stored on the server to ensure that each frame of the image meets the required clarity standard.

[0072] S103: Adjust the video stream based on the resolution, adjust the frame rate of the video stream, set the frame rate to match the dynamic scene changes of the video surveillance, and generate a video standardization processing result;

[0073] In sub-step S103, the frame rate of the video stream after resolution adjustment is adjusted to optimize for dynamic scenes in video surveillance. Frame interpolation technology is used to match the dynamic changes of different scenes. Frame interpolation technology generates intermediate frames by analyzing the motion changes between consecutive frames, improving the smoothness of the video stream. For monitoring fast-moving vehicles and crowds at the entrance, the dynamic scene analysis module calculates and sets the most suitable frame rate. The module adjusts the frame rate according to the scene complexity and the target's movement speed. The frame rate adjustment result directly affects the video surveillance effect and storage requirements. A high frame rate ensures smooth and clear motion, while a low frame rate reduces data storage space. The adjusted video standardization processing result is stored in the central monitoring system for security analysis and future review, ensuring the accuracy and efficiency of the processing.

[0074] Please see Figure 3Based on the video standardization processing results, the steps of detecting objects in the video data by configuring the learning speed to match the dynamic changes in the video, adjusting the size and position of the recognition boxes to reduce the overlap between multiple boxes, and obtaining the target detection box information are as follows:

[0075] S201: Based on the video standardization processing results, adjust the learning rate of the target detection model, match the dynamic changes in the video data, optimize the model's response speed to various scenarios, and generate learning rate adjustment results;

[0076] The specific formula for adjusting the learning rate of the object detection model is as follows:

[0077]

[0078] in, The learning rate adjustment function is used to calculate the current iteration number. The learning rate function incorporates cosine annealing to optimize dynamic adjustment of the learning rate, ensuring a larger learning rate in the early stages of training for rapid convergence, and gradually decreasing it towards the end of training to refine the adjustment of model parameters. This represents the current iteration number, indicating a point in time during the learning process, and is used to adjust the learning rate in real time. The total number of iterations is defined as the preset total number of iteration cycles during the entire training process, used for normalization. The ratio is adjusted to ensure a reasonable distribution of learning rate adjustments throughout the training cycle. This is the minimum learning rate, ensuring that the learning rate does not drop to an extremely low level that is ineffective for model training during the reduction process, thereby maintaining the effectiveness and stability of the method throughout the training process. Pi is a constant introduced in the calculation to ensure that the adjustment of the learning rate has a smooth, periodic transition, adapting to different training phases.

[0079] formula:

[0080]

[0081] Detailed explanation of the formula and its calculation derivation:

[0082] The formula is used to dynamically adjust the learning rate to adapt to the training needs at different stages. It uses the cosine annealing method to gradually reduce the learning rate as the number of iterations increases, stabilize it at a low but non-zero learning rate, and refine the model adjustment in the later stages of training.

[0083] Parameter meanings and settings:

[0084] This represents the current iteration number, assuming it is the 50th iteration, which indicates a specific point in time during the training process.

[0085] This represents the total number of iterations. Assuming the total number of iterations is set to 200, it indicates the length of the training period.

[0086] The minimum learning rate is assumed to be 0.01, ensuring that the learning rate does not drop to 0 at the end of training, thus maintaining the effectiveness and stability of the method during training.

[0087] Pi, approximately 3.14159, is used in the formula as a factor for periodic adjustment.

[0088] Substitute the parameters into the formula to calculate:

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096] Calculation results This indicates that at the 50th iteration, the learning rate was adjusted to 0.1537. This value gradually decreased from the initial stage, which is consistent with the adjustment strategy of cosine annealing. This ensures fine-tuning of the model weights in the later stages of training, avoids overfitting, and retains sufficient learning capacity to continue optimizing model performance. This smooth learning rate reduction strategy helps to achieve more stable and accurate model training results.

[0097] S202: Based on the learning rate adjustment result, detect moving objects in the video frame, record the position and movement information of multiple targets, and generate target position data;

[0098] In sub-step S202, moving objects in the video frame are detected based on the learning rate adjustment result. The process uses a convolutional neural network in deep learning to identify and record the position and movement information of multiple targets within the video frame. Using the YOLO model, multiple targets and their positions in the image are quickly identified in a single viewing. The process includes inputting the adjusted video stream into the model, which calculates the probability distribution and position coordinates of the targets in each frame through forward propagation. Low-probability detection results are filtered out through non-maximum suppression to ensure that the output target position data is accurate and reliable. The position and movement information of each detected target are then encoded into a data packet for tracking analysis and subsequent behavior recognition. The generated target position data is used for real-time monitoring and long-term behavior analysis.

[0099] S203: Based on the target location data, reduce the overlap between boxes by adjusting the size and position of the recognition box, optimize the recognition accuracy and the tracking ability of the target, and generate target detection box information;

[0100] In sub-step S203, based on target location data, the size and position of the recognition boxes are adjusted to reduce overlap between boxes. Using bounding box regression technology in image processing, the recognition boxes are automatically adjusted according to the actual size of the target to ensure that the box matches the target size and minimizes overlap. The process includes analyzing target location data, calculating the spatial distribution around each target, and then dynamically adjusting the size and position of each recognition box. The IoU calculation standard is used to evaluate the coverage efficiency of the recognition boxes before and after adjustment, optimize the recognition boxes, improve the accuracy of target recognition, and enhance the model's ability to track targets. The adjusted recognition box information is used to update the tracking method of the monitoring system, enabling the system to effectively identify and track various dynamic targets. The generated target detection box information is crucial for subsequent security monitoring and behavior analysis.

[0101] Please see Figure 4 Based on the target detection box information, the steps of analyzing the position parameters of the identified target, calculating the position change of the target between consecutive frames, identifying the target's motion path and adding marker information to generate motion path tracking information are as follows:

[0102] S301: Based on the target detection box information, analyze and identify the position coordinates of multiple targets, compare the coordinate data between consecutive frames, analyze the movement direction and speed of the targets in the video, and generate position change analysis results;

[0103] In sub-step S301, the position coordinates of multiple targets based on the target detection box information are analyzed. Optical flow is used to identify and calculate the movement of targets between consecutive video frames. The motion pattern and speed of objects in the image sequence are calculated. The motion speed and direction of a point are estimated by comparing the position change of the same point in two consecutive frames. The process includes extracting the position coordinates of the targets in each frame, using the Lucas-Kanade method to estimate the displacement vector of each target between consecutive frames, and calculating the movement speed and direction. The position change data of each target is accurately calculated and compared to ensure the accuracy of the analysis results. The generated position change analysis results describe the movement trajectory of each target in the video. The information is used for subsequent passenger flow analysis.

[0104] S302: Based on the position change analysis results, track the movement paths of multiple targets in real time, record the movement trajectory and duration of the targets in the video sequence, and generate trajectory recording results;

[0105] In sub-step S302, the movement paths of multiple targets are tracked in real time based on the position change analysis results, and the movement trajectories and durations are recorded in the video sequence. The process employs Kalman filter technology, which is used to estimate the state of a linear dynamic system. In video tracking applications, the future position of the target is predicted by setting the initial state and measurement error, and then corrected based on the actual measurement values ​​provided in each frame. The target position is updated in real time, and the movement trajectory of each target is continuously updated through a method that records the position and timestamp of each appearance of the target to ensure the continuity and accuracy of the trajectory. The generated trajectory recording results provide detailed data support for subsequent analysis, including abnormal behavior detection in security monitoring or customer behavior patterns in retail analysis.

[0106] S303: Based on the trajectory recording results, identification markers are added to multiple targets by associating them with the target's motion trajectory and timestamp, generating movement path tracking information;

[0107] In sub-step S303, based on the trajectory recording results, an identification tag is added to each target by associating it with the target's motion trajectory and timestamp. The process uses a target identification method, which assigns a unique identifier to each target based on its frequency of appearance and motion pattern in the video. This includes analyzing the recorded motion trajectory data, using image recognition technology to identify the characteristics of each target, including color, size, or shape, and combining the data with timestamps to create a continuous tracking tag for each target. This enables the system to continuously track the same target across multiple frames of video, maintaining tracking continuity even when the line of sight is temporarily obstructed. The generated motion path tracking information provides a complete overview of target movement, offering important data resources for applications such as security monitoring and traffic management.

[0108] Please see Figure 5 Based on the movement path tracking information, the specific steps for obtaining the re-identification and classification results are as follows: Feature vectors of multiple targets are extracted and compared with feature vectors already stored in the database to calculate the similarity score between the feature data.

[0109] S401: Based on the movement path tracking information, identify the feature vectors of multiple targets according to the image information, including the target size, geometric shape, and color information, and generate feature vector extraction results;

[0110] In substep S401, feature vectors of multiple targets are extracted based on the movement path tracking information. The process uses convolutional neural networks in deep learning to identify the size, geometry, and color information of the targets. This includes using pre-trained network models such as ResNet or VGGNet to process the input image data. The network extracts features at different levels through a multi-layer structure. The features include basic edge, color, shape, and texture information. The scattered feature information is integrated into a fixed-length feature vector through a fully connected layer. The feature vector of each target is calculated and stored for subsequent feature matching. The generated feature vector extraction results include detailed descriptions of each target, providing a basis for subsequent matching and analysis.

[0111] S402: Based on the feature vector extraction results, the similarity score between the two sets of vectors is calculated by comparing them with the targets identified in the database, and feature matching results are generated;

[0112] The above content utilizes the formula:

[0113]

[0114] Calculate the similarity score between two sets of feature vectors to assess the degree of matching between targets;

[0115] In the formula, Cosine similarity, representing the similarity between two vectors, is used to measure how close the directions of the two vectors are. For vectors sum vector The One element;

[0116] Detailed explanation of the formula and its calculation derivation:

[0117] Suppose there are two eigenvectors and , , Calculate the similarity score:

[0118]

[0119]

[0120]

[0121]

[0122] result This indicates that the two feature vectors have a high degree of similarity, reflecting that the two targets have a high matching probability.

[0123] S403: Based on the feature matching results, by judging the matching status between the target and the known targets in the database, the matched targets are classified and their identity labels are updated, the re-identification status of the targets is identified, and re-identification classification results are generated;

[0124] In sub-step S403, based on the feature matching results, target re-identification and classification are performed. The process uses a support vector machine classifier to process the matching results and update the target's identity label. Each target is compared with known targets in the database according to the similarity score. A threshold is set to determine whether the target matches. If they match, the target is classified using an SVM model. The model predicts the new target classification based on the labels in the training data. The classification results confirm the re-identification status of each target. The information is used to update the target information in the system database to ensure that the data recorded in the system is up-to-date. The generated re-identification and classification results provide accurate target identification and tracking for subsequent monitoring activities and data analysis.

[0125] Please see Figure 6 Based on the re-identification and classification results, combined with the store layout parameter settings for entering the detection buffer, the steps of analyzing the entry and exit behavior of multiple targets in real time, updating and recording the entry and exit status of multiple targets, and generating the area entry detection results are as follows:

[0126] S501: Based on the re-identification and classification results, analyze the layout of the target 4S store and configure the boundary parameters of the entrance buffer zone, including the spatial layout of the store entrance and the direction of pedestrian flow, and generate the buffer zone parameter configuration.

[0127] In sub-step S501, the layout of the 4S store is analyzed based on the re-identification and classification results. The boundary parameters of the entrance buffer are configured. The process uses spatial analysis technology and flow direction model to quantify the spatial layout of the store entrance and the direction of pedestrian flow. The store floor plan is drawn using CAD software and imported into the GIS system. Flow direction model methods, such as flow field simulation, are applied to predict the flow trend of pedestrians under different entrance layouts. The size and shape of the entrance buffer are adjusted according to the simulation results to adapt to the most efficient pedestrian flow distribution and avoid congestion. The buffer parameter configuration includes the boundary length, width and its position relative to the store entrance. The generated buffer parameter configuration results are recorded and used for subsequent entry and exit management strategies to ensure efficient management and safety monitoring of pedestrian flow.

[0128] S502: Based on the buffer parameter configuration, the video data is analyzed in real time, the location data is used to determine and identify targets that cross the buffer, the entry and exit status of the targets is recorded, and a real-time entry and exit status record is generated.

[0129] In sub-step S502, based on the buffer parameter configuration, the system analyzes video data in real time to identify and record the entry and exit status of targets that cross the buffer. It employs the Sobel edge detection method combined with multi-target tracking technology. The configured buffer parameters are imported into the video analysis system. The system receives the video stream in real time and uses the boundary detection method to determine the buffer boundaries. It tracks the movement path of each target using multi-target tracking. When a target crosses the set buffer boundary, the system automatically marks and records its entry and exit status, including entry or exit timestamps. This improves the efficiency of store security management and provides real-time data support for subsequent customer flow analysis. The generated real-time entry and exit status records provide detailed dynamic monitoring results for store management.

[0130] S503: Based on the real-time entry and exit status record, update the target entry and exit data in the database in real time, including the identification code of the target and entry and exit time information, and generate the area entry detection result;

[0131] In sub-step S503, target entry and exit data in the database is updated in real time based on real-time entry and exit status records. Real-time database update technology, including triggers and event-driven programming, is used to synchronize the data detected by the video analytics system. The system identifies the identifier code of each target based on the entry and exit status records obtained from video analytics and captures entry and exit time information. By writing database triggers, the corresponding entries in the database are automatically updated once a new entry or exit event is detected, ensuring the timeliness and accuracy of the database information. This supports the needs of store management and security monitoring systems. The generated area entry detection results are fed back to managers in real time through the database, providing immediate security alerts and customer flow statistics, thereby enhancing the store's operational and monitoring capabilities.

[0132] Please see Figure 7 Based on the area entry detection results, the steps of counting pedestrians entering and exiting the store entrance, recording the characteristics and identification information of multiple pedestrians, assessing the manpower allocation required for the target store's operation, and generating store customer flow statistics are as follows:

[0133] S601: Based on the area entry detection results, calculate the pedestrian flow at the store entrance, including the number of people entering and leaving the store, and generate a pedestrian flow record;

[0134] In sub-step S601, pedestrian traffic at the store entrance is calculated based on the area entry detection results. This involves statistical analysis and data integration. A video analytics system is used to automatically count the number of people entering and leaving the store. Convolutional neural networks are used to detect and distinguish human figures to ensure accuracy. In each video frame, the method identifies pedestrians passing through the store entrance and classifies them as "entering" or "leaving". Each time a pedestrian is detected, the system automatically updates the entrance traffic counter and records a timestamp to achieve real-time traffic monitoring. Historical data is stored in log files for long-term trend analysis. The generated pedestrian traffic records illustrate the number of people entering and leaving each day, providing data support for store operation decisions.

[0135] S602: Based on the pedestrian flow record, record the features and identification information of multiple targets, including appearance features, timestamps, and entry and exit status, to provide data support for pedestrian re-identification and generate pedestrian feature analysis data;

[0136] In sub-step S602, based on pedestrian traffic records, the system records and analyzes the features and identification information of multiple targets. The system uses facial recognition technology and feature extraction methods to identify the appearance features of each pedestrian passing through the store entrance. Using the deep learning model FaceNet, high-dimensional feature vectors are extracted from pedestrian images captured by video. For each identified individual, their appearance features, timestamp, and entry / exit status are recorded and stored in the database to support real-time security monitoring, provide data resources for pedestrian re-identification tasks, and provide reference for subsequent customer behavior analysis and security management.

[0137] S603: Based on the pedestrian feature analysis data, by performing time series analysis on the store's customer traffic data, calculate the changing trend of customer traffic, assess the manpower allocation required for the target store, and generate store customer traffic statistics.

[0138] The specific formula for calculating the trend of passenger flow is as follows:

[0139]

[0140] in, Represents a point in time The trend estimate is used to predict passenger flow at next point in time. Represents a point in time It provides the most accurate and current passenger flow data, reflecting actual passenger volume. Represents a point in time Actual passenger flow, provided slightly earlier Passenger flow information Represents a point in time The actual passenger flow helps in analyzing earlier changes in passenger flow. Represents a point in time The actual passenger flow is used as a reference for long-term trend analysis. yes The weighting coefficients determine the importance of the most recent data in the overall trend analysis. yes The weighting coefficients are used to adjust the contribution of the penultimate period data to the prediction model. yes The weighting coefficients affect the role of the penultimate period data in the overall analysis. yes The weighting coefficients are used to determine the influence of the earliest data in trend prediction.

[0141] formula:

[0142]

[0143] Detailed explanation of the formula and its calculation derivation:

[0144] The formula is used to calculate the predicted customer flow trend value of a store at the current point in time;

[0145] Parameter meanings and settings:

[0146] Indicates a point in time The actual passenger flow data is assumed to be 120, 115, 110, and 105 people.

[0147] The weighting coefficients for each time point are assumed to be 0.4, 0.3, 0.2, and 0.1, reflecting that data closer to the current time point has a greater impact on trend prediction.

[0148] Substitute the parameters into the formula to calculate:

[0149]

[0150]

[0151] The result of 115 people indicates that at that time point... The predicted customer flow is 115 people. This value is calculated based on the actual customer flow data from the last four times and their respective weights. It is used to predict the next customer flow trend and help store management to make reasonable allocations of human resources and resources.

[0152] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for calculating customer flow in 4S stores based on improved pedestrian re-identification, characterized in that, The method includes: Based on image acquisition equipment, video data at the entrance of the 4S store is collected in real time. Combined with the input requirements of video processing, the video data is standardized to generate standardized video processing results. Based on the video standardization processing results, by configuring the learning speed to match the dynamic changes in the video, objects in the video data are detected, and the size and position of the recognition box are adjusted to reduce the overlap between multiple boxes, thereby obtaining target detection box information. Based on the video standardization processing results, the steps of detecting objects in the video data by configuring the learning speed to match the dynamic changes in the video, adjusting the size and position of the recognition boxes to reduce the overlap between multiple boxes, and obtaining the target detection box information are as follows: Based on the video standardization processing results, the learning rate of the target detection model is adjusted to match the dynamic changes in the video data, optimize the model's response speed to various scenarios, and generate learning rate adjustment results. Based on the learning rate adjustment result, moving objects in the video frame are detected, the position and movement information of multiple targets are recorded, and target position data is generated. Based on the target location data, the size and position of the recognition box are adjusted to reduce overlap between boxes, thereby optimizing the recognition accuracy and target tracking ability, and generating target detection box information. Based on the target detection box information, the position parameters of the identified target are analyzed, the position change of the target between consecutive frames is calculated, the motion path of the target is identified and marked information is added, and motion path tracking information is generated. Based on the mobile path tracking information, feature vectors of multiple targets are extracted and compared with feature vectors stored in the database. The similarity score between feature data is calculated to obtain the re-identification and classification result. Based on the re-identification and classification results, combined with the store layout parameter settings, the system enters the detection buffer, analyzes the entry and exit behavior of multiple targets in real time, updates and records the entry and exit status of multiple targets, and generates the area entry detection results. Based on the area entry detection results, the number of pedestrians entering and exiting the store entrance is counted, the characteristics and identification information of multiple pedestrians are recorded, the manpower allocation required for the operation of the target store is assessed, and store customer flow statistics are generated.

2. The improved 4S store customer flow statistics method based on pedestrian re-identification according to claim 1, characterized in that, The video standardization processing results include image resolution adjustment information, frame rate adjustment records, and timestamp records. The target detection box information includes the location recognition box size parameters, recognition box position information, and recognition box location information. The movement path tracking information includes the target's motion speed data, motion direction information, and time series of motion trajectory. The re-identification and classification results include the target's feature similarity score, a list of confirmed targets, and feature library matching information of targets to be re-identified. The region entry detection results include target entry and exit status records, behavior analysis of targets within the buffer zone, and spatial position change records of targets. The store customer flow statistics include the number of counted pedestrians, pedestrian feature and identification datasets, and customer flow change trend data.

3. The improved 4S store customer flow statistics method based on pedestrian re-identification according to claim 1, characterized in that, Based on image acquisition equipment, video data is collected in real time at the entrance of the 4S store. Combined with the input requirements for video processing, the video data is standardized to generate the standardized video processing result. The specific steps are as follows: Based on the image acquisition equipment, the acquisition range is optimized by adjusting the camera's focal length and viewing angle, and video data of the 4S store entrance is acquired in real time. The time information of the acquisition equipment is calibrated to optimize the consistency between the video data and the actual time, and video acquisition results are generated. Based on the video acquisition results, and according to the input requirements of video processing, the resolution of multiple frames in the video is adjusted to optimize image clarity and generate a resolution-adjusted video stream. Based on the resolution, the video stream is adjusted, the frame rate of the video stream is adjusted, the frame rate is set to match the dynamic scene changes of the video surveillance, and a standardized video processing result is generated.

4. The improved 4S store customer flow statistics method based on pedestrian re-identification according to claim 1, characterized in that, The specific formula for adjusting the learning rate of the object detection model is as follows: in, The learning rate adjustment function is used to calculate the current iteration number. The learning rate function incorporates cosine annealing to optimize dynamic adjustment of the learning rate, ensuring a larger learning rate in the early stages of training for rapid convergence, and gradually decreasing it towards the end of training to refine the adjustment of model parameters. This represents the current iteration number, indicating a point in time during the learning process, and is used to adjust the learning rate in real time. The total number of iterations is defined as the preset total number of iteration cycles during the entire training process, used for normalization. The ratio is adjusted to ensure a reasonable distribution of learning rate adjustments throughout the training cycle. This is the minimum learning rate, ensuring that the learning rate does not drop to an extremely low level that is ineffective for model training during the reduction process, thereby maintaining the effectiveness and stability of the method throughout the training process. Pi is a constant introduced in the calculation to ensure that the adjustment of the learning rate has a smooth, periodic transition, adapting to different training phases.

5. The improved 4S store customer flow statistics method based on pedestrian re-identification according to claim 1, characterized in that, Based on the target detection box information, the steps of analyzing the position parameters of the identified target, calculating the position change of the target between consecutive frames, identifying the target's motion path and adding marker information to generate motion path tracking information are as follows: Based on the target detection box information, the position coordinates of multiple targets are analyzed and identified, and the coordinate data are compared between consecutive frames to analyze the movement direction and speed of the targets in the video and generate position change analysis results. Based on the position change analysis results, the movement paths of multiple targets are tracked in real time, and the movement trajectories and durations of the targets in the video sequence are recorded to generate trajectory recording results; Based on the trajectory recording results, identification markers are added to multiple targets by associating them with the target's motion trajectory and timestamp, generating movement path tracking information.

6. The improved 4S store customer flow statistics method based on pedestrian re-identification according to claim 1, characterized in that, Based on the movement path tracking information, the specific steps for extracting feature vectors of multiple targets and comparing them with feature vectors already stored in the database to calculate the similarity score between feature data and obtain the re-identification and classification result are as follows: Based on the movement path tracking information, feature vectors of multiple targets are identified according to image information, including the target size, geometric shape, and color information, and feature vector extraction results are generated. Based on the feature vector extraction results, the similarity score between the two sets of vectors is calculated by comparing them with the targets identified in the database, and feature matching results are generated. Based on the feature matching results, the matching targets are classified and their identity labels are updated by judging the matching status between the target and known targets in the database, the re-identification status of the target is identified, and the re-identification classification result is generated.

7. The improved 4S store customer flow statistics method based on pedestrian re-identification according to claim 1, characterized in that, Based on the re-identification and classification results, combined with the store layout parameter settings for entering the detection buffer, the steps of analyzing the entry and exit behavior of multiple targets in real time, updating and recording the entry and exit status of multiple targets, and generating the area entry detection results are as follows: Based on the re-identification and classification results, the layout of the target 4S store is analyzed, and the boundary parameters of the entrance buffer zone are configured, including the spatial layout of the store entrance and the direction of pedestrian flow, and the buffer zone parameter configuration is generated. Based on the buffer parameter configuration, the video data is analyzed in real time, the location data is used to determine and identify targets that cross the buffer, the entry and exit status of the targets is recorded, and a real-time entry and exit status record is generated. Based on the real-time entry and exit status records, the target entry and exit data in the database is updated in real time, including the identification code of the target and entry and exit time information, and an area entry detection result is generated.

8. The improved 4S store customer flow statistics method based on pedestrian re-identification according to claim 1, characterized in that, Based on the area entry detection results, the steps of counting pedestrians entering and exiting the store entrance, recording the characteristics and identification information of multiple pedestrians, assessing the manpower allocation required for the target store's operation, and generating store customer flow statistics are as follows: Based on the area entry detection results, the pedestrian flow at the store entrance is calculated, including the number of people entering and leaving the store, and a pedestrian flow record is generated. Based on the pedestrian flow records, the characteristics and identification information of multiple targets are recorded, including appearance features, timestamps, and entry and exit status, providing data support for pedestrian re-identification and generating pedestrian feature analysis data; Based on the pedestrian feature analysis data, by performing time series analysis on the store's customer traffic data, the changing trend of customer traffic is calculated, the required manpower allocation for the target store is assessed, and store customer traffic statistics are generated.

9. The improved 4S store customer flow statistics method based on pedestrian re-identification according to claim 8, characterized in that, The specific formula for calculating the trend of passenger flow is as follows: in, Represents a point in time The trend estimate is used to predict passenger flow at next point in time. Represents a point in time It provides the most accurate and current passenger flow data, reflecting actual passenger volume. Represents a point in time Actual passenger flow, provided slightly earlier Passenger flow information Represents a point in time The actual passenger flow helps in analyzing earlier changes in passenger flow. Represents a point in time The actual passenger flow is used as a reference for long-term trend analysis. yes The weighting coefficients determine the importance of the most recent data in the overall trend analysis. yes The weighting coefficients are used to adjust the contribution of the penultimate period data to the prediction model. yes The weighting coefficients affect the role of the penultimate period data in the overall analysis. yes The weighting coefficients are used to determine the influence of the earliest data in trend prediction.

Images