| 王苁蓉.基于改进高斯混合模型的运动目标分割算法[J].轻工机械,2022,40(5):34-42 |
| 基于改进高斯混合模型的运动目标分割算法 |
| Moving Object Segmentation Algorithm Based on Improved Gaussian Mixture Model |
| |
| DOI：10.3969/j.issn.1005 2895.2022.05.006 |
| 中文关键词: 运动目标分割 像素速度 高斯混合模型 前景提取 阴影去除 |
| 英文关键词:moving object segmentation pixel speed Gaussian mixture model foreground extraction shadow removal |
| 基金项目:国家自然科学基金项目(62072416)；国家自然科学基金项目(61873246)。 |
|
| 摘要点击次数: 370 |
| 全文下载次数: 469 |
| 中文摘要: |
| 针对运动目标分割中出现的孔洞等问题，课题组提出了一种基于改进高斯混合模型（Gaussian mixture model，GMM）的运动目标分割算法；针对运动目标分割中出现的阴影和鬼影等问题，提出了一种基于位置约束和加权HSV（hue saturation value，HSV）的阴影检测算法。首先，将运动目标在相邻2帧之间移动的速度作为像素速度，根据像素速度来动态调整每个像素点的学习率，实现背景模型的动态更新，从而有效避免前景中出现孔洞等干扰；其次，提出一种基于位置约束和加权HSV颜色模型的特征向量构建方法，使用余弦相似性进行相关性分析，避免了分割过程中出现阴影干扰的现象。实验结果表明：该算法能够准确地分割出运动目标，目标完整率、阴影检测率和判别率分别达到95.11%、96.18%和96.45%。与其他方法相比，运动目标分割的准确率提高了2.09%，运动目标分割性能良好。 |
| 英文摘要: |
| Aiming at the problems of holes in the segmentation of moving objects, a moving object segmentation algorithm based on improved Gaussian mixture model was proposed. Aiming at the problems of shadows and ghosts in the segmentation of moving objects, a shadow detection algorithm based on location constraints and weighted HSV was proposed. Firstly, the speed of the moving target between two adjacent frames was defined as the pixel speed, and the learning rate of each pixel was dynamically adjusted according to the pixel speed to realize the dynamic update of the background model, so as to effectively solve the interference problems such as holes in the foreground; Secondly, a feature vector construction method was proposed that fuses the location constraints and the weighted HSV color model, and cosine similarity was used for correlation analysis to avoid the phenomenon of shadow interference during segmentation. The experimental results show that the proposed algorithm can accurately segment the moving objects. The target integrity rate, shadow detection rate and discrimination rate can reach 95.11%, 96.18% and 96.45% respectively. Compared with other methods, the accuracy of moving target segmentation is improved by 2.09%, and the performance of moving target segmentation is good. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
