High-Sensity Pedestrian Detection from Aerial View

High-density pedestrian detection from aerial view is a novel and challenging work in both the field of computer vision and pedestrian flow research. Although many pedestrian detection algorithms have emerged, their performance are not good enough, especially in high-density situations. Therefore, we modified the original YOLOv5 model, and make it more suitable for this job. Based on the video data recorded by unmanned aerial vehicle (UAV), we build the pedestrian dataset, and use the modified YOLOv5 model to train the heads and shoulders of pedestrians in the dataset. The new detection results show that our proposed model can finish the job of high-density pedestrian detection from aerial view very well.