Mobile-Yolo: A Lightweight Pedestrian Detection Model for Advanced Driver Assistance Systems

The pedestrian detection system (PDS) is a function of an advanced driver assistance system (ADAS), which aims to locate pedestrians and help drivers avoid accidents. To ensure this, a fast and accurate pedestrian detection model is required, which should be lightweight enough to adapt to mobile vehicles. The target detection algorithm Yolov4 performs well in terms of detection speed and accuracy, but its algorithm is too large to be applied to PDS. In this study, a lightweight pedestrian detection model, Mobile-Yolo, is proposed, which is based on Yolov4 and combined with the depth separable convolution of Mobilenet. Channel pruning method is utilized for model compression. Greatly reducing parameter size, Mobile-Yolo can balance detection accuracy and speed in detecting pedestrians. The evaluation of Mobile-Yolo resulted in 85.57% of AP and an inference speed of 19.1 FPS. The results are basically equivalent to that of Yolov4 for the same data set.