Image Based Visual Servoing for Landmine Detection using Quadrotors

This paper presents a technical approach for landmine detection with quadrotors using image based visual servoing (IBVS) on thermal images. Considering the difference in temperature between the ground soil and possible buried targets, thermal images are used to estimate target position in an unscented Kalman filter framework: accelerometers, gyroscopes and GPS measurements are integrated in loosely coupled manner as the navigation solution. Al first, the quadrolor quipped with single thermal camera estimates the target's depth by following a special ellipsoid trajectory and then using IBVS, it approaches to the target up to one meter distance. In order to control the drone to flight close to the ground, a ground effect compensated controller is also considered in the implementation. Results show that for landmine detection purposes using IBVS on thermal images, with ground effect compensated control, an overall error around 20 pixels is achievable.