Unmanned Aerial Vehicle (UAV) Based Running Person Detection from a Real-Time Moving Camera*

Abstract

The information captured by Unmanned Aerial Vehicles (UAVs) are highly exploited to detect a running person which is given in this paper. In this scheme, 11 frame per seconds and an adequate detection precision in outdoor background was realized, using one dispensation thread without resorting to distinct hardware. The high precision and realtime detection were made promising by two aids. First, we used a progression of preprocessing procedures to extract the regions of interest (ROI), this includes spatial domain analysis, calculating the optical flow in every two consecutive images and having a predefined threshold built on optical flow to select actual areas as ROIs. Secondly, we also used relatively minor-batch models to train our 5-layer convolutional neural networks (CNN) in order to realize an adequate detection ratio. The experiments from numerous videos shot in diverse time and locations proved that the proposed scheme can detect running person in outdoors efficiently and enhance the Realtime necessity with a very high detection ratio.