Review of Typical Vehicle Detection Algorithms Based on Deep Learning
Gaoyan Dong *
School of Information Engineering, North China University of Water Resources and Electric Power, Zhengzhou - 450000, China.
Bin Li
School of Information Engineering, North China University of Water Resources and Electric Power, Zhengzhou - 450000, China.
Yiliang Chen
School of Information Engineering, North China University of Water Resources and Electric Power, Zhengzhou - 450000, China.
Xinyu Wang
School of Information Engineering, North China University of Water Resources and Electric Power, Zhengzhou - 450000, China.
*Author to whom correspondence should be addressed.
Abstract
Object detection is the crucial task in the field of computer vision. In recent years, intelligent driving technology and intelligent transportation system have set off a boom. Therefore, vehicle object detection has also become a hot research task in the field of computer vision and deep learning. With the rapid development of deep learning, the current mainstream vehicle detection algorithms are Convolutional Neural Networks (CNN)-based two-stage and one-stage object detection algorithms. Because of the local nature of the image presented by CNN, the global receptive field of the network is limited. At the same time, Transformer shows a strong long-distance dependence characteristic, and opens up a new idea of combining images with Transformer. Therefore, the research of object detection algorithm based on Transformer gradually causes a boom. This paper mainly introduces the advantages and disadvantages of several representative algorithm models, and makes a summary and prospect.
Keywords: Vehicle detection, deep learning, convolutional neural networks, transformer