Remote Sensing based Human Monitoring and Recognition over Public Surveillance

Due to the ability to capture data from such distinct aerial perspectives, UAVs have gained numerous applications in drone imaging, human action recognition (HAR), search and rescue (SAR), and surveillance. However, such applications have consequences such as complex backgrounds and partial occlusions. This research presents a unique technique to tackle these issues for HAR using drone video data. We present an approach beginning with frame extraction from drone footage and preprocessing. First, we perform gamma correction and then we conduct background subtraction. YOLO detects humans, and key point extraction is achieved by extracting the important human joint points. The features were extracted, including full body features acquired by SIFT and LBP, key point features via spin images, and the Unique Shape Context (USC). Next, we improve these features using linear discriminant analysis (LDA), followed by a neuro-fuzzy classifier. The accuracy was assessed on the UAV Human and Drone Action datasets with scores of 56.4% and 86.7%, respectively.