AUTOMATIC TRACKING OF COMPLEX OBJECTS UNDER NATURAL CONDITIONS

We deal with the problem of tracking freely flying insects that have been filmed under natural conditions. Although this appears to be an easy task for human observers it proves to be quite difficult to implement in computer vision systems. The main goal of our project was to automatize the reconstruction of animal flightpaths in a single-frame video film analysis. In addition to standard image-processing tools we employed a number of physical constraints as well as general knowledge to produce a system with acceptable performance. In this article we describe basic algorithms and techniques that provide a pragmatic solution in a specific case but that could in principle be applied to a large class of related problems. Our system copes with up to 5 degrees of freedom and a number of other variables such as brightness, illumination, object shape, occlusion and light reflection. We discuss the performance of a first implementation as well as remaining problems.