A Review of Vehicle Detection Methods Based on Computer Vision

Abstract

With the increasing number of vehicles, there has been an unprecedented pressure on the operation and maintenance of intelligent transportation systems and transportation infrastructure. In order to achieve faster and more accurate identification of traffic vehicles, computer vision and deep learning technology play a vital role and have made significant advancements. This study summarizes the current research status, latest findings, and future development trends of traditional detection algorithms and deep learning-based detection algorithms. Among the detection algorithms based on deep learning, this study focuses on the representative convolutional neural network models. Specifically, it examines the two-stage and one-stage detection algorithms, which have been extensively utilized in the field of intelligent transportation systems. Compared to traditional detection algorithms, deep learning-based detection algorithms can achieve higher accuracy and efficiency. The single-stage detection algorithm is more efficient for real-time detection, while the two-stage detection algorithm is more accurate than the single-stage detection algorithm. In the follow-up research, it is important to consider the balance between detection efficiency and detection accuracy. Additionally, vehicle missed detection and false detection in complex scenes, such as bad weather and vehicle overlap, should be taken into account. This will ensure better application of the research findings in engineering practice.