Recurrent YOLO and LSTM-based IR single pedestrian tracking
Sungmin Yun, Sungho Kim · 2019
In this paper, we develop a new approach of spatially supervised recurrent convolutional neural networks for thermal infrared (TIR) visual pedestrian tracking. Our method extends the YOLO deep convolutional neural network into the spatiotemporal domain using Long Short-Term Memory (LSTM). In particular, we propose a new CNN model that satisfies both accurate and robust while maintaing low computational cost in the TIR-VOT. Our experimental results and performance comparison with state-of-the-art tracking methods on challenging benchmark video tracking datasets. The proposed TIR-ROLO method processes images in real-time at 45 fps and achieves the best tracking performance in occlusion scenario.