Motion Detection by Classification of Local Structures in Airborne Thermal Videos

In this paper we present a method to efficiently detect moving objects namely vehicles in airborne thermal videos. The motion of the sensor is estimated from the optical flow using projective planar homographies as transformation model. A three level classification process is proposed: On the first level we extract interest location applying the Foerstner operator. These are subject to the second finer level of classification. Here we distinguish four classes: 1. Vehicles cues; 2. L-junctions and other proper fixed structure 3. Tjunctions and other risky fixed structure. 4. A rejection class containing all other locations. This classification is based on local features in the single images. Only structures from the L-junctions class are traced as correspondences through subsequent frames. Based on these the global optical flow is estimated that is caused by the platform movement. The flow is restricted to planar projective homographies which highly reduces the computational time. This opens the way for the third classification. The vehicle class is refined using motion as feature. Inconsistency with the estimated flow is a strong evidence for movement in the scene. This is done by computing a difference image between two sequent frames transformed by a homography to be taken from the same position. The difference images are pre-processed using vehicle properties and velocity.