YOLO v5 for SDSB Distant Tiny Object Detection
Xiaodong Yu, Ta Wen Kuan, Yuhan Zhang, Taijun Yan · 2022
To reach SDSB (Self-Driving Sweeping Bot) in an efficient-sweeping manner, data collection of visual images regarding sweeping target must be conducted prior to analyze the required sweeping objects with other noises. In this work, three categorized target objects including, fallen leaves, speed bumps and manhole cover etc. are involved in training and validation phases. To reach SDSB with real-time object detection, the work further investigated one-stage of Yolo v5 learning approach of four version including, Yolo v5s, Yolo v5m, Yolo v51, and Yolo v5x, wherein Yolo v5s in terms of its benefits on lower frame rate, high-accuracy, and high-speed characteristics for real-time object detection. Furthermore, to detailed analyze the Yolo v5s performance on training and validation set, several indices including, box loss, objectness loss, classification loss, precision, recall and mean average precision (mAP) in terms of epoch number, are also reported in the experiments.