Moving targets detection from UAV based on a robust real-time image registration algorithm

In this paper, a real time method for image registration dedicated to small moving target detection (MTD) from a moving camera is proposed. The main application is the road surveillance from UAV. The algorithm must work under strong constraints: real time performances for VGA images at 30 fps with a standard PC and shaking camera with irregular movements. Given the medium-altitude of the UAV, the estimated transformation is a similarity. Our proposed algorithm is an hybrid method using an ESM algorithm (Efficient Second-Order Minimization) and Harris corners matching, with pyramidal implementation. An M-Estimator is used at each loop of the ESM in order to be robust to moving objects, occlusions and camera noise. We have tested our method on UAV highway images and this coupling allows for robustness, low time computation and accuracy. Even if the camera may have important oscillations, we get a high precision alignment of the images and thus a reliable detection of the moving targets.