Advanced Object Tracking in Video Surveillance Systems with Adaptive Deep SORT Enhancement

Object tracking is a crucial feature of video surveillance systems that are essential for maintaining awareness and detecting potential threats. Advanced solutions are needed to overcome the obstacles associated with video object tracking, including the complexity of everyday environments and the massive amount of data. Traditional tracking algorithms often struggle with the complexity of dynamic situations, necessitating the use of deep learning methods. This paper presents an innovative deep learning-based object tracking system that uses Multi-Level Glow-Worm Swarm Convolution Neural Networks (MLGS-CNNs) to detect objects in video frames. Subsequent object tracking is facilitated by the adaptive Deep Simple Online Real-time Tracking (DeepSORT) algorithm by incorporating an optimized Kalman filter instead of a conventional Kalman filter. The Waterwheel Plant Optimization (WPO) method is used to tune the noise covariances of the Kalman filter to further improve the tracking accuracy. Comprehensive performance criteria, including metrics such as Multiple Object Tracking Accuracy (MOTA), Multiple Object Tracking Precision (MOTP), Integrated Detection and False-alarm Rate (IDF1), Mostly Tracked (MT), and Mostly Lost (ML), are used to evaluate the effectiveness of our method.