A student trajectory estimation method based on campus road network and adaptive multi-travel mode

Abstract

This invention discloses a student trajectory prediction method based on campus road networks and adaptable to multiple travel modes, belonging to the field of smart campus technology. The method comprises the following modules: S1, Target Student Image Information Input Module: Inputs the image information of the target student; S2, Target Student Image Information Capture Module: Acquires student image information within a target time period through campus monitoring equipment and outputs capture records; S3, Target Student Auxiliary Feature Extraction Module: Extracts facial, clothing color, and height range features from the captured images to obtain valid records; S4, Multi-Feature Fusion Similarity Matching Analysis Module: Uses multi-feature fusion similarity matching to filter out valid capture sequences of the target student; S5, Target Student Trajectory Analysis Module: Generates a complete trajectory through dynamic road network weight modeling and a trajectory completion algorithm with spatiotemporal constraints. This invention possesses the technical advantages of adapting to multiple travel modes on campus, high identity recognition accuracy, accurate trajectory prediction, and close alignment with the campus scenario.

Claims

1. A method for predicting student trajectories based on campus road networks and adapting to multiple travel modes, characterized in that, Includes the following steps: S1, Target Student Image Information Input Module: Inputs the image information of the target student and transmits it to the video monitoring equipment; S2, Target Student Image Information Capture Module: Acquires student image information within a target time period through campus monitoring equipment and outputs capture records; S3, Target Student Auxiliary Feature Extraction Module: Extracts facial features, clothing color, and height range features from the captured image to obtain effective records; S4. Multi-feature fusion similarity matching analysis module: Uses multi-feature fusion similarity matching to filter out the effective capture sequences of the target students; S5, Target Student Trajectory Analysis Module: Generates a complete trajectory through dynamic road network weight modeling and a trajectory completion algorithm with spatiotemporal constraints, including: Dynamic road network weight modeling: Construct a campus road network graph G=(V,E). For three travel modes—walking, shared bicycles, and electric vehicles—calculate the travel time of road segments by combining slope and peak-hour corrections. Map video surveillance locations to road network nodes, and obtain effective capture sequences Seq based on time-location continuity. valid ; Trajectory completion with spatiotemporal constraints: for valid capture sequences Seq valid The algorithm uses an improved A* algorithm to search for paths that satisfy the time difference constraint among adjacent capture records. After ranking the candidate paths by their credibility, the most likely path is output. The effective capture points and the completed path are then integrated to generate a complete trajectory.

2. The student trajectory prediction method based on campus road network and adaptable to multiple travel modes as described in claim 1, characterized in that: In step S1, the image information includes the student's face photo, upper body clothing color, lower body clothing color, height range, and time period of appearance.

3. The student trajectory prediction method based on campus road network and adaptable to multiple travel modes as described in claim 1, characterized in that: In step S2, the output capture record Record=(StudentFace,Cam ID,Time,Picture), where StudentFace is the captured student face image, Cam ID is the capture camera number, Time is the capture time, and Picture is other captured student images.

4. The student trajectory prediction method based on campus road network and adaptable to multiple travel modes as described in claim 1, characterized in that: In step S3, valid records are retained and formatted uniformly as follows: Data=(StudentFace,CamID,Time,F uppercolor F lowercolor，Fheight Among them, F uppercolor F is the feature vector of the upper body clothing color. lowercolor为下身衣着颜色特征向量，Fheight The clothing color feature vector is extracted as follows, using the geometric proportions of the human body when standing / walking to segment the upper and lower body regions. The segmented upper and lower body region images are then converted from RGB space to HSV space, and the color histogram is statistically analyzed and normalized into a vector F. uppercolor∈[0,1] 64 Flowercolor∈[0,1] 64 The height range encoding value Fheight is estimated by estimating the height range and encoding it as a numerical value based on the camera installation height, shooting angle, and the pixel ratio of the human body in the image.

5. The student trajectory prediction method based on campus road network and adaptable to multiple travel modes as described in claim 1, characterized in that: In step S4, the input target student image information is used as a seed template to construct a standard feature Template=(F StudentFace T F uppercolor T F lowercolorT Fheight T)，其中：F StudentFaceT Fuppercolor is used to measure the facial features of the target students. T F is the mean vector of the histogram of upper body colors. lowercolor T F is the mean vector of the histogram of lower body colors. height T Calculate Data for each valid snapshot based on the average height range. i The similarity Sim(Datai, Template) of the multi-feature fusion with the standard feature Template is used to filter out valid capture records with a similarity ≥ a preset threshold.

6. The student trajectory prediction method based on campus road network and adaptable to multiple travel modes as described in claim 5, characterized in that: The multi-feature fusion similarity Sim(Data) i The template is calculated using the weighted summation formula: Sim = w1 × Sim uppercolor +w2×Sim lowercolor+w3 ×Sim height +w4×Sim StudentFace +w5×Sim time , where: Sim StudentFace For facial similarity, Sim uppercolorcolor For the similarity of upper body color characteristics, Sim lowercolor为下身颜色特征相似度，Simheight为身高相似度， Sim height =1-∣F height,i -F height T | / 0.5, when |F height,i -F height T When |>0.5m, Sim height =0.

7. The student trajectory estimation method based on campus road network and adaptable to multiple travel modes as described in claim 1, characterized in that: In step S5, the calculation method for road segment travel time in the dynamic road network weight modeling is as follows: for walking mode: t ewalk = / (v walkmax ×k_walk)×C Bicycle mode: t ebike = / (v bikemax ×k_bike)×C Electric vehicle mode: t eev = / (v evmax ×k_ev)×C, where L(Ei) is the length of road segment Ei, v walkmax v bikemax v evmax K represents the base speed for walking, cycling, and electric vehicles, respectively. k_walk, k_bike, and k_ev are the slope correction coefficients for the corresponding modes of transportation, and C is the congestion coefficient.

8. The student trajectory prediction method based on campus road network and adaptable to multiple travel modes as described in claim 7, characterized in that: The congestion coefficient C is set as follows: C=1.5 during peak school hours and C=1.0 during off-peak hours. During peak school hours, the travel time is multiplied by 0.8 based on the slope correction speed. The time-location continuity filtering method is as follows: After sorting the captured records by time, calculate the capture time difference ΔT between adjacent records and the road network travel time difference ΔTe. If ΔT is within [ΔT...], then... e ×0.8, ΔT e If the similarity is within the range of [×1.2], then two records are retained; otherwise, records with low similarity are discarded.

9. The student trajectory prediction method based on campus road network and adaptable to multiple travel modes as described in claim 1, characterized in that: In step S5, during the trajectory completion process under spatiotemporal constraints, the improved A* algorithm uses the capture time difference ΔT as the core constraint and the travel time after slope correction as the path cost function to search for paths that satisfy te_min ≤ ΔT ≤ te_max, where te_min is the shortest travel time and te_max is the longest reasonable travel time. The basis for ranking the credibility of the multiple paths is the weighted sum of the overlap rate between the path and the student's historical travel trajectory, the completeness of road segment monitoring coverage, and the slope adaptability.

Images