Real-time tracking-by-detection framework for traffic applications via deep learning based convolutional neural network

Abstract

Vision-based target tracking is one of the core parts of intelligent traffic video surveillance systems due to its assistance in reducing risks of traffic accidents and traffic jams on roads. This paper proposes a new tracking-by-detection method in the domain of traffic applications by using the powerful ability of deep learning based object detection technique into vision-based tracking for vehicles. The proposed method uses transfer learning technique to achieve state-ofthe-art tracking performance but building upon a powerful object detector while only requiring few hundreds of images data for training. The experimental results not only validates the performance of the proposed transfer learning technique and also shows that tracking can be achieved using this approach. Furthermore, qualitative and quantitative results on challenging dataset show that the proposed tracking method achieves competitive performance with the state-of-the-art methods. © JES 2020 on-line.