Hybrid approach for suspicious object surveillance using video clips and UAV images in cloud-IoT-based computing environment

In broad terms, it can be said that the existing video surveillance systems with automatic face recognition and manual face detection in crewless aerial vehicles (UAVs') video frames did not have recognition accuracy above 90%. It is because of using a limited amount of Eigenfaces for the principal component analysis (PCA) transform. Detecting a face in a cloud-IoT video frame involves extricating video/image windows into two basic classes: one class will contain faces (training the surroundings) and the other will contain matching (in the foreground). The problem of face detection is further convoluted by using inconsistent video/image qualities, lighting conditions, and geometries, and including the options of inequitable occlusion and disguise. Therefore, an ultimate face detection technique in video frames/images would be intelligent enough to identify the existence of any face in any set of foreground/background and lighting conditions. The authors further found that the implementation of a completely automated iris image-based face recognition and detection system can be useful for observation purposes like ATM (automated teller machine) user security whereas the proposed and implemented automated face recognition using UAVs video frames in cloud-IoT integrated distributive computing is ideal for mug-shot matching and surveillance of suspicious objects. It is because of the presence of controlled conditions while gathering mug shots. The proposed hybrid face detection/object detection-based video surveillance system was tested under very robust conditions in a cloud-IoT (Internet of Things) integrated distributive computing environment and it is envisaged in the experimental study that the real-world performance of the proposed hybrid system will have far better accuracy than the existing systems. The authors observed during experiments that the proposed cloud-IoT integrated hybrid approach for video surveillance of suspicious objects in a distributed real-time environment has high accuracy, low (overall error rate), and very high (average recall rate) for self-generated real-time datasets and benchmark datasets of KTH, COCO Detection, ImageNet, Kaggle, Vlomonaco Github, Pascal VOC, UCF-11, AVA, Collective Action, MSRA-B, SegTrack, ViSal, SegTrack V2, VOS, VOS-N, and DAVIS. The results of the proposed cloud-IoT integrated video surveillance system re-insure that the choices selected by the authors are reliable, efficient, and robust. Further, these results are adequate for the surveillance of suspicious objects using UAV video clips/images in a cloud-IoT integrated environment but further improvements in results are always possible and enviable.