Region Proposal Approach for Human Detection on Aerial Imagery

Abstract

In this paper we evaluated region proposal based CNN approach in human body detection from aerial perspective. Particular emphasis is on the automation of the detection for supporting search and rescue missions. This challenging task is characterized by two important requirements. The first requirement for proposed algorithm is real-time speed of execution and the other is exceptional detection quality on complex natural environment images. Evaluation is performed on high spatial resolution images with high level of details that were collected by UAV platforms. Evaluated method based on FasterRCNN detection model showed promising preliminary results, as well as fast processing of high resolution images. Overall, detection model achieved 88.3% recall with precision of 67.3%.