Fine-grained vehicle recognition under low light conditions using EfficientNet and image enhancement on LiDAR point cloud data
Guanqiang Ruan, Tao Hu, Chenglin Ding, Kuo Yang, Fanhao Kong, Jinrun Cheng, Rong Yan
Abstract
The detection and recognition of vehicles are crucial components of environmental perception in autonomous driving. Commonly used sensors include cameras and LiDAR. The performance of camera-based data collection is susceptible to environmental interference, whereas LiDAR, while unaffected by lighting conditions, can only achieve coarse-grained vehicle classification. This study introduces a novel method for fine-grained vehicle model recognition using LiDAR in low-light conditions. The approach involves collecting vehicle model data with LiDAR, performing projection transformation, enhancing the data using contrast limited adaptive histogram equalization combined with Gamma correction, and implementing vehicle model recognition based on EfficientNet. Experimental results demonstrate that the proposed method achieves an accuracy of 98.88% in fine-grained vehicle model recognition and an F1-score of 98.86%, showcasing excellent performance.