Human Detection With Fisheye Camera

Most of the human detection and tracking studies in camera images are done by means of cameras with normal viewing angles. Detection and tracking in images of overhead cameras with fisheye angle of view involve various difficulties. Due to the wide field of view cameras with overhead view providing a 360 degrees angle of view, the shapes of the objects on the edge of the image are distorted. Detection algorithms have difficulty detecting deformed objects. In this study, it is aimed to obtain a model suitable for fish-eye cameras by training human samples taken from fish-eye camera images with CNN algorithm and to measure the performance of the model with YOLO detection algorithm. A simple interface design has been made to facilitate the tracking of the image frames from the YOLO detection algorithm.