A Temporal-spatial Background Model for Video Objects Detection

Foreground objects detection from sequence images is the foundation of intelligence surveillance system. The key step of background subtraction based detection method is building background of the scene. A novel temporal-spatial background of single modal scene has been proposed in this paper. Firstly, the temporal Gaussian model of each pixel in background is built. Then it clusters the models by their mean value to get together pixels in the same region of scene and it separates pixels which have been put together but not connected spatially. So that it generates a background which is more accurate to the scene than the pixel based statistical model. After building the background, it updates the pixels of one region synchronously. That improves the speed of model updating and restrains the noise in the homogeneous regions of background. Experimental results show that the integrity of detected objects with the proposed background is better than the temporal Gaussian model background, and achieves satisfactory performance while detected the objects that move in the scene and stop.