Human Detection in Aerial Images using Deep Learning Techniques

Activity recognition in drone-based surveillance is related to many computer vision problems such as pose estimation, object detection, image retrieval, face recognition, frame tagging in videos, and video action recognition. In a drone-based surveillance system, detection and recognition of human activities in a single frame is a challenging task as the clips are shot from an aerial view. Unlike activity recognition in static camera-captured videos where spatio-temporal features are utilized, they are not utilized in drone-captured images. This problem is addressed in this paper using HOG and Mask-RCNN. Experimental results show that the proposed method can be obtained more accurate results in many drone-based frames. This work produces high-quality segmentation through instance segmentation in addition to the histograms gradient-based method and also improves the accuracy of object detection in aerial images and gives the best classification results.