基于混合高斯模型的新型目标检测系统

针对传统混合高斯模型计算量过大及其在非平稳背景下存在的问题，提出一种新型运动目标检测系统。该系统引入模型等权重初始化策略，改善了视频检测初始阶段的效果；通过基于线性均差的模型匹配方法，减少了对方差的运算次数，有效减少了模型的计算量；加入干扰信息处理模块，以增强模型在复杂背景下的生存能力。经实验验证，新型系统准确检测出了视频初始阶段的多运动目标，可使进入场景后停留的目标快速融入背景，并能有效克服非平稳背景的扰动。实验结果表明该系统相比经典模型，准确性和鲁棒性均有明显改善。

New objects detection system based on mixture Gaussian model

The traditional mixture Gaussian models need much computation and have difficulty in detecting the moving objects under a non-stationary background. Therefore，a new object detection system based on mixture Gaussian model was proposed. An equal weight initial strategy was introduced in the system to improve the results of video detection in the initial stage. Meanwhile, the operation times of variance were reduced by a model matching method based to the linear mean deviation, the calculation of the models was reduced effectively, and the survival ability of model in complex background was enhanced by adding a disturbance information processing module. The experimental results show that the new system can accurately detect multiple moving objects in the initial video stage, and can fast amalgamate the entering and remaining objects into the background. Besides, it can effectively overcome the disturbance of nonstationary background. The research clearly shows that the accuracy and robustness of system are significantly improved.