Knowledge discovery of suspicious objects using hybrid approach with video clips and UAV images in distributed environments: a novel approach

The current video surveillance systems that employ manual face detection and automatic face recognition in unmanned aerial vehicles (UAVs) have limited accuracy, typically below 90%. This is due to the utilization of a small number of Eigenfaces for principal component analysis transformation. Detecting faces in cloud-based Internet of Things (IoT) video frames involves separating video/image windows into two classes: one with faces (to train the surroundings) and the other with matches (in the foreground). The face detection process is further complicated by geometries, inconsistent image/video qualities, and lighting conditions, as well as the possibility of partial occlusion and disguises. Moreover, a fully automated iris image-based face recognition and detection system could prove useful in surveillance applications such as automated teller machine user security, whereas an automated face recognition system using UAV video frames in a cloud-integrated-IoT-based distributed computing environment is better suited for mug-shot matching and surveillance of distrustful objects. This is because controlled conditions are present when capturing mug shots. The proposed hybrid approach was rigorously tested, and the experimental results suggest that its real-world performance will be far more accurate than existing systems. Intelligent surveillance knowledge databases contain vast amounts of information on landmarks, terrain, events, activities, and entities that need to be efficiently and accurately processed and disseminated. Therefore, discovering the appropriate knowledge to detect distrustful objects plays a crucial role in future analysis. The experimental findings indicate that the proposed hybrid approach has high accuracy, low overall and average error rates, and very high average recall rates for benchmark and self-generated datasets. These results demonstrate the robustness, efficiency, and reliability of the authors’ choices. Although further improvements in results are possible, the proposed approach is sufficient for detecting distrustful objects.