A vehicle re-identification method based on multi-task joint discriminative learning

Abstract

The invention discloses a vehicle re-identification method for multi-task joint discrimination learning. The method uses a multi-branch network to jointly learn multiple tasks to obtain the fine-grained discriminative features of the vehicle. The network obtains the network output feature vector through the two branches of attribute learning and ID learning, and at the same time uses a metric learning and ID learning task to constrain the feature vector, and obtains more robust features through joint learning of these four tasks. Among them, ID learning uses ArcFL loss function different from other methods, and metric learning uses Trihard loss function different from other methods. Through the proposal of an innovative network structure and the improvement of the loss function, the accuracy of vehicle re-identification and retrieval has been significantly improved. The invention is realized based on a large vehicle data set of a road monitoring scene, and can be effectively applied to a vehicle search task.

Description

technical field [0001] The invention belongs to the fields of image processing, computer vision and pattern recognition, and in particular relates to a vehicle re-recognition method for multi-task joint discrimination learning. Background technique [0002] Vehicles, as important objects in urban traffic scenarios, have attracted a lot of attention in the field of computer vision research in recent years. Vehicle re-identification technology is an important research content of intelligent transportation systems. In terms of intelligent management and safety maintenance, in the face of automatic toll collection, searching for specific vehicles and other scenarios, vehicle re-identification tasks need to be completed. Common vehicle re-identification methods are usually based on metric learning and a method of combining vehicle model learning and metric learning. The method based on metric learning, such as the method proposed by Zhang et al. (refer to Zhang’s method: Zhang Y...

AI Technical Summary

Problems solved by technology

However, the network structure of this method is simple, and the fine-grained features of the vehicle cannot be fully learned, and the accuracy of vehicle re-identification is not high enough.

Images