Research on Key Technologies of Object Detection in Autonomous Driving Environments Based on Deep Learning

Autonomous driving demands object detection with high accuracy and real-time performance to assist environmental perception by identifying pedestrians, vehicles, traffic signs, and more. However, challenges remain in balancing speed and accuracy, as well as handling complex lighting conditions and object occlusion. This study explores commonly used object detection methods in autonomous driving, including two-stage detectors and single-stage detectors. It further analyzes how feature extraction, multi-scale detection, lightweight networks, and data augmentation techniques enhance model performance. Through optimization strategies such as model compression, pruning, and quantization, the study achieves a balance between model accuracy and detection speed. Experimental results demonstrate that the optimized model achieves high real-time performance and accuracy on datasets like COCO and KITTI, with superior results in terms of mean Average Precision (mAP) and Frames Per Second (FPS) compared to unoptimized models, showcasing its potential for autonomous driving applications.