Aiming at the problem of low detection accuracy caused by occlusion in real-time target detection of wild animal, considering the special time-series relationship of video, this paper proposed a method of wild animal occlusion detection based on entropy and motion offset. The method performed feature extraction on targets by YOLOv3 model, and obtained the detection position and category of the occlusion object to be detected according to the image corresponding to the lowest information entropy of the video sequence. The experimental results show that the real-time object detection method combining YOLOv3, information entropy, time series relationship and position offset can solve the problem of wild animal occlusion detection well and improve the stability and accuracy of real-time object detection.

[2019-12-05] Printed Article /doc/2018.08.0587/v2