基于超像素信息反馈的视觉背景提取算法

发布时间：2018-11-24 08:52

【摘要】：针对经典视觉背景提取算法长时间存在鬼影、动态背景导致的高频噪声以及背景模型误更新等问题,提出一种改进的视觉背景提取算法。该算法将原始图像分割为若干个超像素区域,在超像素分割区域,对视觉背景提取算法检测结果进行像素点再分类,在目标检测的初始阶段实现鬼影信息的准确检测,并更新鬼影区域像素点的背景模型,从根本上解决了全局范围内鬼影检测的难题。根据运动目标的超像素对前景目标内的空洞进行快速纠正,实现前景目标的小范围填补,同时完成对背景超像素内高频噪声的检测和滤波,并增强检测结果的稳健性。利用数据集进行的测试实验结果表明,与传统算法相比较,该算法的精确率和识别率等指标均显著提高。
[Abstract]:An improved visual background extraction algorithm is proposed to solve the problems of ghost image, high frequency noise caused by dynamic background and false updating of background model in classical visual background extraction algorithm for a long time. The algorithm divides the original image into several super-pixel regions. In the hyperpixel segmentation region, the detection results of the visual background extraction algorithm are subdivided into pixels, and the accurate detection of ghost image information is realized in the initial stage of target detection. The background model of pixels in the ghost region is updated to solve the problem of global ghost detection. According to the super-pixel of the moving target, the holes in the foreground target can be corrected quickly, and the small range of foreground target can be filled. At the same time, the detection and filtering of the high-frequency noise in the background super-pixel can be completed, and the robustness of the detection result will be enhanced. The experimental results using data sets show that the accuracy rate and recognition rate of the algorithm are significantly improved compared with the traditional algorithm.
【作者单位】：河北工业大学控制科学与工程学院;
【基金】：国家自然科学基金青年基金(61403119) 河北省自然科学基金青年基金(F2014202166) 天津市科技特派员项目(15JCTPJC55500) 天津市智能机器人重大专项(14ZCDZGX00803)
【分类号】：TP391.41

【相似文献】