IBWO-Optimized YOLOv5 Framework with Keyframe Selection for Real-Time Surveillance Object Detection

Video surveillance plays a crucial role in ensuring public safety, with object detection being a fundamental task in identifying potential threats. Traditional object detection techniques often struggle with efficiency and accuracy due to redundant data and computational constraints. This paper proposes an efficient framework for object detection in surveillance videos by integrating advanced deep learning and optimization techniques. The process begins with converting surveillance video into individual frames, followed by keyframe selection using a distance-based measure to minimize redundancy. The selected keyframes are then processed using the YOLOv5 deep learning model, known for its high accuracy and real-time detection capabilities. To enhance performance, an Iterative Beluga Whale Optimization (IBWO) algorithm is incorporated for optimizing the loss function of YOLOv5. This integration ensures improved detection accuracy and computational efficiency. Experimental results demonstrate the effectiveness of the proposed approach, making it a robust solution for real-time surveillance applications. The outcomes of the proposed method are inspected on an Abandoned Objects Dataset. The results obtained from the proposed method developed in the scope of this research are 99.01% accuracy and 98.46% specificity respectively. The proposed method outperforms existing methodologies in real-time surveillance object detection.