ABSTRACT
In recent times, object tracking has become an essential process effectively utilized in surveillance, medical imaging, video analytics, Augmented Reality, traffic monitoring and other areas. Performing real-time tracking in daylight is relatively easy than that performed during the night due to the better quality of illumination during the day. Thermal infrared object tracking works beyond the illumination constraint, monitoring the object of interest in conditions of complete darkness. The tracking efficacy of such systems depends massively on the quality of the acquired video stream and also how the video stream is processed. With the emergence of deep learning, the thermal imagery associated with deep neural networks have offered the most promising tracking results. This chapter aims to review the object tracking frameworks based on deep learning for infrared videos. The major categories are: (i) review on detection-based and detection-free thermal tracking techniques; (ii) review on single and multiple thermal objects tracking techniques; and (iii) review on object fusion based thermal tracking techniques. Apart from these, a review on the benchmark thermal data sets have also been presented and the conventional as well as he state-of the-art evaluation metrics have been discussed. The review accesses the gap amid the reported research and futuristic demands for infrared object tracking.
