Improvement of YOLOv3 network based on ROI

The paper adopts the YOLOv3 network, by removing the calculation of non-interest areas to improve the operation speed of YOLOv3. It increases the operating speed in proportion to the size of the non-interest area. The author studies the characteristics of the convolutional layer and the YOLO layer, and proposes a new processing method from the experimental results. In this paper, the ROI image processing technology in the video detection sequence is used to filter the ROI region, eliminate the calculation outside the ROI, and improve the detection speed and accuracy of the YOLOv3 network.