Learning motion detectors by genetic programming

Motion detection in videos is a challenging problem that is essential in video surveillance, traffic monitoring and robot vision systems. In this paper, we present a learning method based on Genetic Programming(GP) to evolve motion detection programs. This method eliminates the need for pre-processing of input data and minimizes the need for human expertise, which are usually critical in traditional approaches. The applicability of the GP-based method is demonstrated on different scenarios from real world environments. The evolved programs can not only locate moving objects but are also able to differentiate between interesting and uninteresting motion. Furthermore, it is able to handle variations like moving camera platforms, lighting condition changes, and cross-domain applications.