An analysis of various deep learning-based target detection algorithms in the field of autonomous driving

Abstract

Target detection is a crucial research objective within the domain of computer vision, finding extensive applications in areas such as robotics, autonomous driving, industrial inspections, and various other fields. Based on the foundation of deep learning theory, this paper systematically summarizes the application and prospect of each type of target detection algorithm (based on regression and based on candidate region) on automatic driving, compares the advantages and disadvantages of the two types of algorithms, as well as the results of detecting traffic signals, traffic vehicles, and pedestrians, and focuses on the application scenarios as well as the comparison of advantages and disadvantages of each method. A systematic summary of the current development results is made. Among them, the most prominent target detection in the field of transportation is undoubtedly the algorithms of various branches of the YOLO series.