Ground Target Automatic Recognition and Tracking in Complex Battlefield Environment

With the developments of science and the changes of modern warfare modes, the precise guidance and attack towards hostile targets in complex battlefield environments has extremely important military needs and research significance. However, the existing guidance mode is often constrained by factors such as battlefield background interference, target appearance variation and limited data links, which in turn affects tracking and strike performance. To this end, we present an integrated recognition and tracking scheme for the ground targets in complex battlefield environment. Specifically, with the missile-target distance information, we design an adaptive scale estimation scheme to tackle the aspect ratio variation during tracking. Furthermore, an efficient tracking monitoring index is advocated to judge the tracking condition as well as forecast the potential tracking failure. Once tracking failure is determined, a deep-learning based detection module is activated to re-detect the target as well as correct the tracking drift. In the strike stage of the mission, the key region of the hostile target would be identified and locked by the missile in order to maximize the attack effect. The proposed recognition and tracking scheme are tested upon extensive actual videos and achieved a tracking accuracy over 85%, which demonstrates its effectiveness as well as provides a novel solution for accurate and robust attack on enemy targets in a complex battlefield environment.