Multi-Object Tracking framework with YOLOv9 Detector and DeepSORT Algorithm based on Generalized Intersection over Union (GIOU)

Multi-object tracking (MOT) systems are crucial in various real-life applications, including smart video surveillance, traffic monitoring, robotics, sports analytics, customer behavior analysis, and drone control. Despite their widespread use, MOT faces persistent challenges, such as occlusion, ID switching, variations in lighting, background interference, and high object similarity. Deep learning methods like Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and DeepSORT are commonly employed to address these issues. This research proposes a Multi-Object Tracking framework by combination of the multi-object detector, YOLOv9 and a DeepSORT algorithm based on Generalized Intersection over Union (GIoU) metric to tackle ID switching and occlusion challenges, ultimately improving multi-object tracking accuracy on MOT 16 datasets. Results of the proposed system are compared with the results of the DeepSORT algorithm based on Intersection over Union (IoU) metric. According to the experiment result, the proposed system can track multi-objects more accurately than the DeepSORT based on IoU metric.