Target Detection and Tracking in Infrared Videos Using Frequency Domain Analysis and Machine Learning for Surveillance

Military surveillance during pitch dark night conditions is of prime importance for the security of any nation. Medium Wave Infra Red (3-5μm) or Long Wave Infra Red (8-12μm) thermal cameras are used in the border areas for reconnaissance. Automatic target detection and tracking in these Infra Red (IR) videos is very important for alertness and preaction. This paper presents an algorithm for detecting and tracking a target in IR video. IR image characteristics are utilized for the detection of target using Fast Fourier Transform, followed by Boosted Butterworth high pass filter to sharpen the image and increase the contrast between the target and its surroundings. Then proper thresholding and blob analysis is performed to detect the target. Then target is tracked using centroid matching function. Machine learning concept is used to develop trajectory prediction algorithm with dynamic indexing of weight of rate of change in angle with respect to the horizontal axis to handle both linear and non-linear trajectories. Experimental results show that the developed algorithm is robust for target detection and tracking in IR videos and is capable of handling human as well as vehicle targets moving in linear or non-linear fashion. The algorithm successfully predicts the target even if the target is occluded partially or fnlly over several frames.