Escape vehicle re-identification method

Abstract

The invention discloses an escape vehicle re-identification method. The method comprises the following steps: (1) constructing a target camera topological network, and predicting an associated cameratrack; (2) carrying out metric learning based on visual angle perception, and learning two different depth metrics in S-view same-visual-angle and D-view cross-visual-angle samples; and (3) performingvehicle re-identification based on a dual-path adaptive attention, wherein dual paths include a global path and a local path, dual-path vehicle re-identification is performed in the S-view same-visual-angle and D-view cross-visual-angle feature spaces respectively in the step (2), global features of the picture are extracted by the global path, and the global features are supplemented by the local path. Through constructing a suspicious vehicle camera topological network, a key monitoring area with an optimal time sequence is acquired; and different loss functions are applied by utilizing deep metric learning, an adaptive attention model is added, a re-identification task is carried out, and a walking track of the vehicle is obtained so that the re-identification accuracy of the escapingvehicle is improved.

Description

technical field [0001] The invention relates to a vehicle re-identification method, in particular to a re-identification method of an escaped vehicle. Background technique [0002] With the development of science and technology and the improvement of people's living standards, the frequency and occupancy of automobiles have gradually increased, and people's awareness of traffic safety and solutions have also increased accordingly. Once a traffic accident occurs, how to contact artificial intelligence identification to quickly and accurately deal with the standardized and intelligent traffic accident is very important. [0003] In recent years, with the introduction of large datasets and the development of deep learning algorithms, as well as the widespread application of traffic cameras, vehicle re-identification based on deep learning has achieved remarkable success in the past decade. Vehicle re-identification technology has great application potential in the fields of ur...

AI Technical Summary

Problems solved by technology

This work faces the following major challenges. First, appearance-based methods often fail to achieve satisfactory results because the differences between different vehicle classes taken from similar viewpoints are small, while the intra-class differences of the same vehicle taken from different viewpoints are huge.

Although deep metric learning has achieved some success in the research of feature acquisition with changing viewing angles, it is still very challenging for extreme viewing angle changes (such as 180°) of vehicles. The impact is very obvious

Second, some subtle clues such as tire models, window stickers, window frames and some custom decorations in the car are difficult to obtain from the global features in appearance detection.

Images