Tracking and Classification of Multiple Human Objects Directly in Compressive Measurement Domain for Low Quality Optical Videos

Data collected in compressive measurement domain can save data storage and transmission costs. In this paper, we summarize new results in human target tracking and classification using compressive measurements directly. Two deep learning algorithms known as You Only Look Once (YOLO) and residual network (ResNet) have been applied. YOLO was used for object detection and tracking and ResNet was used for human classification. Extensive experiments using low quality and long range optical videos in the SENSIAC database showed that the proposed approach is promising.