Robust method of detecting moving objects in videos evolved by genetic programming

Abstract

In this paper we investigated the use of Genetic Programming (GP) to evolve programs which could detect moving objects in videos. Two main approaches under the paradigm were proposed and investigated, single-frame approach and multi-frame approach. The former is based on analyzing individual video frames and treat them independently while the latter approach consider a sequence of frames. In the single-frame approach, three methods are investigated including using pixel intensity, pixel hue value and feature values. The experiments on Robosoccer field show that GP could detect the target under different lighting conditions and could even handle arbitrary camera positions. Although there was no domain knowledge had been provided during evolution, GP was able to produce moving object detectors that were robust and fast.

Detecting motion from noisy scenes using Genetic Programming

“…The GP-based motion detection method is introduced by Fang et al [10] and was proved effective in various real world scenarios [11]. In this paper, we further study the effectiveness of this method by extending it to deal with additive noise and other variations.…”

Evolving frame splitters by Genetic Programming

“…Trujillo and Olague [13] successfully used GP to mark interest points in images, which helped identify edges of objects and changes in colour and illumination. Fang, Pinto and Song used GP to evolve motion detection programs which can not only differentiate moving targets from uninteresting moving objects but also handle unstable background [8], [12].…”