Motion detection in complex environments by genetic programming
Brian Alphonse Pinto, Andy Song · 2009
Detecting motions is an important aspect of machine vision. However real world vision tasks often contain interfering motion information which is not of interest. To tackle this difficult task, we adapted Genetic Programming into this domain. The GP-based methodology presented in this paper does not require the implementation of existing motion detection algorithms. The evolved programs can detect genuine moving objects such as cars and boats, while ignoring background movements such as waving trees, rippling water surface and even pedestrians. These programs provide reliable performance under different lighting conditions, either indoors and outdoors. Furthermore no preprocessing of video input is required which is usually mandatory in conventional vision approaches.