Target Tracking in Blind Range of Radars With Deep Learning

Surveillance radars form the first line of defense in border areas. But due to highly uneven terrains, there are pockets of vulnerability for the enemy to move undetected till they are in the blind range of the radar. This class of targets are termed the `pop up' targets. They pose a serious threat as they can inflict severe damage to life and property. Blind ranges occur by way of design in pulsed radars. To minimize the blind range problem, multistatic radar configuration or dual pulse transmission methods were proposed. Multistatic radar configuration is highly hardware intensive and dual pulse transmission could only reduce the blind range, not eliminate it. In this work we propose, elimination of blind range using deep learning based video tracking for mono static surveillance radars. Since radars operate in deploy and forget mode, visual system must also operate in a similar way for added advantage. Deep Learning paved way for automatic target detection and classification. However, a deep learning architecture is inherently not capable of tracking because of frame to frame independence in processing. To overcome this limitation, we use prior information from past detections to establish frame to frame correlation and predict future positions of target using a method inspired from CFAR in a parallel channel for target tracking.