A vehicle cross-camera tracking method and system based on a twin network

By combining Siamese networks with Kalman filtering and directed connected graphs, multi-dimensional behavioral features of vehicles are extracted, solving the problem of poor robustness in vehicle identification during cross-camera tracking. This achieves highly accurate and robust vehicle identity association, improving the efficiency of intelligent traffic management and security monitoring.

CN122415686APending Publication Date: 2026-07-17SHANGHAI GENTEK CORP LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
SHANGHAI GENTEK CORP LTD
Filing Date
2026-04-22
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing vehicle cross-camera tracking methods have poor robustness in complex monitoring environments and struggle to use vehicle movement patterns and camera topology to determine identity consistency, leading to vehicle identity changes or loss when switching between cameras.

Method used

By employing a twin network combined with Kalman filtering and directed connected graphs, multi-dimensional behavioral features are extracted by constructing vehicle speed fluctuation curves and topological behavior association graphs. Feature encoding is performed using wavelet transform and graph convolution operations to generate behavioral consistency scores to achieve identity association.

Benefits of technology

It significantly improves the accuracy and robustness of vehicle tracking across cameras, reduces identity switching and loss caused by environmental interference, and enhances the computing efficiency and automation of intelligent traffic management and security monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122415686A_ABST
    Figure CN122415686A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of computer vision, and discloses a vehicle cross-camera tracking method and system based on a twin network, which comprises the following steps: collecting a track segment of a target vehicle under a camera; performing Kalman filtering on the track segment to obtain a motion state sequence, performing time series interpolation on the motion state sequence to obtain a speed fluctuation curve; constructing a directed connected graph based on the layout characteristics of the camera, mapping the speed fluctuation curve to the directed connected graph to obtain a topological behavior correlation graph of the target vehicle; jointly encoding the speed fluctuation curve and the topological behavior correlation graph to obtain a behavior feature vector of the target vehicle, and inputting the behavior feature vector into a twin network to obtain a behavior consistency score of the target vehicle; and associating the identity of the target vehicle based on the behavior consistency score, and determining that the target vehicle is the same target vehicle if the behavior consistency score is greater than a preset association threshold; and the application can improve the accuracy and robustness of vehicle cross-camera tracking.
Need to check novelty before this filing date? Find Prior Art