Evaluating Real-Time Object Detection Models for Autonomous Vehicle Vision

Object detection has made remarkable advancement in recent years particularly through CNN-based algorithms, essential for autonomous vehicles. These vehicles need efficient detection systems that balance speed and accuracy in complex environments. The study evaluates several real-time object detection models, such as YOLOv10, YOLOv11, EfficientDet, Faster R-CNN, and SSD focusing on both single-stage and two-stage detectors. Using the Berkeley DeepDrive 100K image dataset and iRain dataset, the study assesses the performance of these models in identifying key road-related objects. Additionally, the paper introduces an enhanced object detection system designed to improve robustness and accuracy in challenging conditions. The results emphasize the crucial role these advanced models have in improving the environmental awareness of autonomous vehicles, leading to better safety and operational performance on the road. Experimental data demonstrate that the proposed system surpasses current models, providing a scalable and dependable solution to boost the safety and efficiency of autonomous systems in real-world settings.