Histogram-based adaptive learning for background modelling: moving object detection in video surveillance

Abstract

The detection of moving object in the presence of complex or cluttered background is a very critical challenge. The moving object may be a person, patient, vehicle, animal or any tissue inside body in medical domain. In this context, this work has proposed a robust background subtraction method for resolving illumination variation and motion-based problems. Initially, this work has developed a background modelling method using initial few frames in training stage. In testing stage, a foreground modelling method is investigated that is able to detect moving object in video frames. In testing stage, this work classify moving pixel with a suitable threshold and update the background using appropriate learning rate. The learning rate is updated through histogram of classified resulting frame and background model. Finally, morphological filters and image processing techniques are applied to improve the detection quality. The employed method also demonstrates how it can be improved using adaptive learning rate-based controlling scheme and the incorporation of feedback-based model updating scheme. It clearly depicts strength of proposed method in handling illumination variation problems and also eliminating moving environmental effects. This method presents significant performance in comparison with considered state-of-the-art methods.