A change detection method for image sequences based on physical models

Change detection methods detect changes between two frames in an image sequence. Fundamental technique for detecting changes uses the difference image between two frames. A change of each pixel is detected if a difference value exceeds pre-set thresholds, which is determined based on an estimated value of variance of noise on images. Not only the noise on images but also illumination changes between frames is one of critical problems for change detection. A recently proposed approach shows better results by using thresholds derived from average of difference image over areas which are estimated as non-change parts. However, such a threshold may not be appropriate since this approach uses no physical parameters such as light sources, reflection of objects. This paper proposes a new change detection method based on physical model, which describes physical parameters such as light sources and reflection of objects, known as illumination model. First, we show derivation of threshold based on illumination model. Threshold is derived from angle of light of sources, gray level of background objects, and normal-vector of background objects. Next, change detection algorithm using such a threshold is shown. Finally, we show experimental results and comparison, in which proposed method improves accuracy of detection results, compared to the conventional method.