Online Multi-Object Tracking Using Convolutional Neural Networks and the Invasive Weed Optimization Algorithm

Multi-object tracking (MOT) is a fundamental problem in computer vision with critical applications in areas such as video surveillance, human-computer interaction, autonomous driving, and video analytics. The main objective of MOT is to estimate the motion trajectories of multiple objects across sequential video frames while preserving their consistent identities throughout the sequence. MOT algorithms are generally categorized into two types: online methods, which process each frame sequentially and make tracking decisions in real-time, and offline methods, which process the entire video or segments of it as a batch to improve accuracy. In this study, we propose an online multi-object tracking method based on convolutional neural networks (CNNs). Unlike traditional approaches with fixed architectures, our method dynamically optimizes the number of hidden layers in the ANN using the Invasive Weed Optimization (IWO) algorithm, a nature-inspired metaheuristic optimization technique. This optimization aims to minimize the classification error, thereby enhancing the tracking performance by selecting a network architecture that is best suited to the complexity of the input data. The proposed system is evaluated using the VS-PETS 2009 benchmark dataset, a widely used dataset for evaluating object tracking algorithms. All simulations and model training are carried out in the MATLAB environment. The experimental results indicate that the proposed method achieves superior tracking accuracy and identity preservation performance compared to conventional tracking methods, demonstrating the effectiveness of combining ANNs with IWO in real-time multi-object tracking scenarios.