Vehicle Detection and Classification Algorithm Based on Improved Mask R-CNN

Abstract

Vehicle detection and classification technology is an important part of ITS system. It is often used in traffic flow statistics, ETC non-parking charging system and other scenarios, which solves a series of problems effectively in traffic control. In this paper, we design ResNet-HDC feature extraction network and New-FPN feature fusion network to address coarse-grained vehicle classification and recognition problem. We improve Mask R-CNN algorithm and propose a vehicle detection and classification algorithm based on improved Mask R-CNN to complete accurate recognition and classification of vehicles. To further optimize the training results, vehicle images are preprocessed before the algorithm model training and high-performance Tesla T4-GPU processor is used to enhance image recognition during the training process. After training and testing, the results show that the method enhance the precision of vehicle detection effectively, and the average accuracy is increased by about 10%.