Track-RNN: Joint Detection and Tracking Using Recurrent Neural Networks

As deep neural networks revolutionize many fundamental computer vision problems, there have not been many works using neural networks to track objects. In this project, we design and implement a tracking pipeline using convolutional neural networks and recurrent neural networks. Our model can handle detection and tracking jointly using appearance and motion features. We use MOT data challenge as a highly occluded single object tracking dataset. We demonstrate good qualitative and quantitative results of our model and discuss how to extend the pipeline to multi-object tracking.