基于自适应混合高斯模型的前景目标检测

摘要

在计算机视觉研究中，从视频序列中提取出前景目标是关键步骤之一。而混合高斯背景模型是前景目标检测的一种常用算法。针对传统混合高斯建模过程中分别对每个像素建立固定个数的高斯模型和相同的学习率这一缺陷，本文先对视频帧进行了分块处理，然后自适应的对每个像素块采取不同的高斯分布个数和学习率，并且在建模过程的不同时间段采用不同的学习率，最后对检测结果在空域上进行数学形态学的处理。实验结果表明，与传统检测方法相比，该方法能够更加准确和快速地检测出前景目标。%The extraction of foreground object from a video sequence is one of the key steps in computer vision re-search. Gaussian mixture model is a kind of commonly-used foreground object detection algorithm. In view of the traditional Gaussian background model process to establish a fixed number of Gaussian model and the same learn-ing rate separately for each pixel,the video frame is divided into blocks firstly,then an adaptive different number of Gaussian distribution and different learning rate for each pixel block is taken at different times by using different modeling learning rate,finally the mathematical morphology for image post-processing in the space domain was ap-plied. The experimental results show that,compared with the traditional detection method,this method has the char-acteristics of quickness and accuracy,thus obtaining better prospects target detection results.