Real-time Pedestrian Detection for Autonomous Driving

Fast and efficient pedestrian detection technology has become an increasingly important task in the autonomous driving technology. Traditional pedestrian detection algorithms are usually too time-consuming to meet the real-time requirements of autonomous driving. In this paper, we propose a new pedestrian detection algorithm based on the Darknet with optimized feature concatenation, hard negative mining, multi-scale training, model pretraining, and proper calibration of key parameters. In view of the characteristics of pedestrian detection system, we adopt the priori experience about the feature box sizes, instead of K-mean clustering algorithm. We also conduct statistical analysis on the dataset pedestrian label, and design the initial value of the pre-selection box that is more in line with pedestrian characteristics. The proposed algorithm not only improves the detection accuracy, but also enhances the efficiency of pedestrian detection. Experimental results on traffic record benchmark demonstrate that the optimized algorithm satisfies the real-time and accuracy requirements of the low-speed autonomous driving.