Object Detection and Tracking in Thermal Video Using Directed Acyclic Graph (DAG)

This paper suggests an incipient approach to perform target detection as well as tracking for single and multiple moving objects in thermal video sequences. Thermal imaging is complimentary to visible imaging as it has capability to detect object in low light or dark conditions by detecting the infrared radiation of an object and creating an image which contains temperature information. The extracted regions are then used for performing the segmentation of targets in thermal videos. In projected method first, Directed Acyclic Graph (DAG) is used for segmentation in thermal videos. Second, to enlarge the set of target proposals, DAG is initialized with an incremented object proposal set in which, from adjacent frames motion based predictions are used. Last, in this paper for selection of the specific object motion scoring function is used, which is having high optical flow gradient between the edges of the object and background is presented. After segmentation of object, centroid based object tracking is performed to track the objects in thermal videos. The proposed method is evaluated on different thermal videos and found to be robust compared with standard background subtraction method.