Detecting human class objects in 5K resolution video who are very far from the camera. Usually humans very close to camera are detected with good precision, while humans away from the cameraget undetected. This is mostly noticed in football, cricket and other games where games are intentionally recorded in Go-Pro/Ordinary cameras to avoid camera cost. We have retrained the Yolo-v3 model with datasets of humans very far from the camera and also conducted experiemnts to understnad the effect of network size during object detection.
Python detect_players