P rogress in T arget D etection T echnology for R oad T raffic V ehicles

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

With the rapid implementation of intelligent transportation systems and autonomous driving technology, vehicle object detection, as the core technology of environmental perception, is widely used in road monitoring, traffic flow statistics, violation identification, and on-board forward perception. Real road traffic scenarios present problems such as large vehicle scale variation, mutual occlusion of targets, complex illumination conditions, and severe weather interference, which can easily lead to missed detection and false detection. This paper divides deep learning-based vehicle detection algorithms into three technical routes: two-stage detection, one-stage detection, and Transformer end-to-end detection, and systematically reviews the core principles, technical evolution paths, and scenario adaptation characteristics of each algorithm. Taking the lightweight YOLO11-n model as the research carrier, a comparative experiment is designed based on the SODA10M large-scale autonomous driving public dataset to quantitatively verify the improvement effect of the multi-scale data augmentation strategy on vehicle detection performance, and multi-angle result analysis is completed using statistical charts, precision curves, and feature heatmaps. Finally, the core challenges faced by current vehicle detection technology are summarized, and future optimization directions are discussed, which can provide theoretical reference for the selection and optimization of intelligent transportation perception equipment. Experimental results show that the multi-scale augmentation strategy improves the mAP@0.5 of YOLO11-n from 74.3% to 78.9% on the SODA10M dataset, demonstrating its effectiveness in addressing vehicle scale variation.