Extended faster R-CNN for long distance human detection: Finding pedestrians in UAV images
Tong Liu, Hui Fu, Qing Wen, Deng Kui Zhang, Ling Fei Li · 2018
Recently, using consumer Unmanned Aerial Vehicles(UAV) for aerial photography has became a trend. However, the images captured from the UAV raise a challenge to the existing pedestrian detection algorithms, because the humans in the image are too blur and too low-resolution resulted from the long distance between the UAV and pedestrians. The problem of detecting long distance humans in an image has always been over-looked, so even the performance of the state-of-the-art detection algorithms are not satisfactory when used on UAV pedestrian detection. In this paper, we extend Faster R-CNN algorithm by proposing an improved Region Proposal Network(RPN) and utilizing object context information to improve the detection performance. The experimental results show that the extended algorithm improves the performance of detecting pedestrians captured by UAV.