Human detection in videos plays an important role in various real life applications. Most of traditional approaches depend on\nutilizing handcrafted features which are problem-dependent and optimal for specific tasks.Moreover, they are highly susceptible to\ndynamical events such as illumination changes, camera jitter, and variations in object sizes. On the other hand, the proposed feature\nlearning approaches are cheaper and easier because highly abstract and discriminative features can be produced automatically\nwithout the need of expert knowledge. In this paper, we utilize automatic feature learning methods which combine optical flow\nand three different deep models (i.e., supervised convolutional neural network (S-CNN), pretrained CNN feature extractor, and\nhierarchical extreme learning machine) for human detection in videos captured using a nonstatic camera on an aerial platform\nwith varying altitudes. The models are trained and tested on the publicly available and highly challenging UCF-ARG aerial dataset.\nThe comparison between these models in terms of training, testing accuracy, and learning speed is analyzed. The performance\nevaluation considers five human actions (digging, waving, throwing, walking, and running). Experimental results demonstrated\nthat the proposed methods are successful for human detection task. Pretrained CNN produces an average accuracy of 98.09%.\nS-CNN produces an average accuracy of 95.6% with soft-max and 91.7% with Support Vector Machines (SVM). H-ELM has an\naverage accuracy of 95.9%. Using a normal Central Processing Unit (CPU), H-ELM�s training time takes 445 seconds. Learning in\nS-CNN takes 770 seconds with a high performance Graphical Processing Unit (GPU).
Loading....