Real Time Multi-Object Tracking based on Faster RCNN and Improved Deep Appearance Metric

Abstract

Computer Vision has set a new trend in image resolution, object detection, object tracking, and more by incorporating advanced techniques from Artificial Intelligence (AI). Object detection and tracking have many use cases such as driverless cars, security systems, patient monitoring, and so on. Various methods have been proposed to overcome the challenges such as long-term occlusion, identity switching, and fragmentation in real-time multi-object detection and tracking. However, reducing the number of identity switches and fragmentation remains unclear in multi-object detection and tracking. Hence, in this paper, we proposed a multi-object detection and tracking technique that involves two stages. The first stage helps to detect the multiple objects with high uniqueness using Faster RCNN and the second stage, Improved Sqrt cosine similarity, helps to track the multiple objects by using appearance and motion features. Finally, we evaluated our proposed technique using the MultiObject Tracking (MOT) benchmark dataset with current state-of-the-art methods. The proposed technique resulted in enhanced accuracy and reduces identity switching and fragmentation.