Research And Improvement Of Motion Target Detection Algorithm For Single Gaussian Background Model

With the development of computer vision and artificial intelligence, people pay more attention to study smart surveillance. Now, artificial intelligence is used in public place and important place. Artificial intelligence realizes four functions:object detection, object recognition, object tracking and object judgment. Above these parts, object detection is the base of all; it affects directly the processing result of related parts. So, many researchers pay deeply research in this field, and propose many effective algorithms.Now, there are four kinds of classic moving object detection algorithms:background model, difference image, kernel density estimate and optical flow. This paper pays attention to three kinds of algorithms:background model, difference image, kernel density estimate in the static sence. And aiming at the non-adaptive and low detectivity problems of single Gaussian background model (SGM), an improved single Gaussian background model method for moving objects detecting is proposed. This method combines SGM and mean shift to detect moving objects. The initial background model is decided by using N frame images, and the moving objects are firstly detected, then the pixels which belong to background points are updated according to the SGM algorithm, and the pixels which do not belong to the background points in the updated background model are corrected using mean shift, last the background model corrected by mean shift is used as the last background model. Lastly the moving objects are detected using the background difference method. Experiments show that the improved method can overcome the non-adaptive shortcoming and have high detectivity.