Visual obstacle detection for UAV

Abstract

The advancement in the Unmanned Aerial Vehicle (UAV) technology has resulted in the increasing usage of automatic applications, particularly with computer vision such as object and human detection. Detecting the presence of humans from a long range through an aerial view is still a challenge to-date. From the view of a UAV, humans are difficult to detect due to having only a small number of pixels. In many applications, human shadows are deemed to be a hassle in image processing and thus removed by many existing tools and algorithms. However, this project would leverage on these shadows to improve the accuracy of human detection. This report would describe and analyze three experiments that would focus on human detection without shadows, with shadows and real-time video human detection. Based on the analysis of the results obtained, it was observed that the accuracy was indeed improved with the usage of shadows.