Application and Development of DETR and Its Variants in Traffic Vehicle Detection

School of Computer Science, Beijing Information Science and Technology University, 102200, Beijing, China

^* Corresponding author’s email: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

With the development of intelligent transportation and autonomous driving, higher requirements are placed on the accuracy, robustness, and real-time performance of vehicle detection in traffic scenarios. Traditional CNN based methods have shortcomings in global relationship modeling and prior dependencies. DETR, as the first end-to-end Transformer detection framework, provides a new path for vehicle detection. However, the original DETR still has limitations in training efficiency, small object detection, and high-resolution computational overhead, which are more prominent in complex traffic scenarios. In response to the above issues, researchers have proposed methods such as multi-scale feature enhancement, lightweight and real-time improvement to enhance detection performance and inference efficiency. The article further explores the research direction of integrating geometric priors, multimodal information, and adaptive reasoning to enhance model robustness and cross scene adaptability. I hope this review can provide reference for the research of intelligent transportation perception systems and promote further application and breakthroughs of Transformer based detection technology in the field of vehicles.