基于多层码书模型的运动阴影检测算法

在实际监控场景中,运动阴影的存在将对目标检测的准确性产生直接影响.针对此问题,本文提出了一种采用三层码书模型以此来提高阴影检测识别率的算法.该算法首先通过传统码书模型获取前景,然后对前景部分建立备选码书模型,再从备选码书中提取出具备阴影特质的点,构建阴影码书模型,最后通过该模型去除运动阴影.实验结果表明,与传统算法相比,该算法对阴影检测识别率有较大的提高.同时,通过对不同场景的对比,结果说明该算法具备良好的鲁棒性.

Movingcast shadow detection by multilayer codebook models

In actual surveillance scenes, shadows severely affect the accuracy of moving objects detection. In this paper, a three layer codebook model is proposed to improve the shadow discrimination rate. Firstly, the foreground points are obtained by the traditional codebook model. Secondly, a cache model is constructed and then the foreground points which meet the characteristics of shadow are selected to form a codebook shadow model. At last, the moving cast shadows are removed through the codebook shadow model. Experimental results illustrate that our method has a superior performance than the conventional shadow detection algorithm. The results obtained with scenes of different types and of different conditions show the robustness of the approach.