Research On Moving Object Detection Algorithms Based On Gaussian Mixture Model

Moving object detection was the foundation research of computer vision, and finding out moving region in the video quickly and availably was the promise of later study, such as moving object tracking and behavior understanding. With the development of study and application, many solutions were proposed to detect moving region from any vision stably, accurately, real-timely, and the most used methods were optical flow, time difference, background subtraction.Because of flexibility in method of modeling and adaptability with the vision, background subtraction became the basic method for motion detection, and most researches in the same area based on background subtraction. Due to simplicity, flexibility, efficiency, Gaussian mixture model was the most popular model for imitating background to be subtracted by corresponding image and was called classic method.On the foundation of background subtraction with adaptive Gaussian mixture model, this paper proposed following improvement to enhance the performance of detection.Used PCNN's regional correlation of segmentation to decide which model is matched with pixel, and changed the single threshold in PCNN's iteration proceeding to multiple thresholds in order to improve the accuracy of matching, and iteration time was decided by simple maximum entropy principle. The improvement enhanced the performance of noise restrain and motion detection result.For the updating of Gaussian mixture model's parameters, a new learning rate was proposed to improve the algorithm's efficiency and stability, and to make the imitated background more accurately reflect the true background.Experiment results proved that this paper's algorithm improved the performance of background noise restrain, motion detection and system stability. The later study could benefit from the improvement.The paper designed an intelligent surveillance system on PC with the theoretic foundation of improved moving object detection algorithm. The system could monitor the designated region efficiently, and could be competent for complex surveillance task.