| 引用本文: | 李乐,张茂军,张文琪,等.联合特征累积分析的压缩域运动对象检测.[J].国防科技大学学报,2012,34(6):54-60.[点击复制] |
| LI Le,ZHANG Maojun,ZHANG Wenqi,et al.Associated features accumulation and analysis for motion object detection in compressed domain[J].Journal of National University of Defense Technology,2012,34(6):54-60[点击复制] |
|
| |
|
|
| 本文已被：浏览 8737次 下载 6156次 |
| 联合特征累积分析的压缩域运动对象检测 |
| 李乐1,2, 张茂军1, 张文琪3, 李永乐1 |
|
(1.国防科技大学 信息系统与管理学院，湖南 长沙 410073;2.
2.武警工程大学 信息工程系，陕西 西安 710086;3.酒泉卫星发射中心，甘肃 酒泉 732750)
|
| 摘要: |
| 依据H.264压缩域中能够反映景物运动变化的MV和DCT系数特征，本文提出了一种多特征联合累积分析的压缩域运动对象检测方法。该方法对压缩码流中各宏块的运动信息进行时空域滤波，并使用雅克比矩阵描述全局运动参数和宏块MV之间的关系，简化参数求解过程，通过比较局部运动和全局运动之间的差异初步检测运动对象；选取宏块周围可靠的运动特征用于宏块DCT系数能量的投影累积，并通过熵能原理在压缩域中选取各个宏块的自适应阈值，检测运动对象的边缘及纹理显著区域；通过一定的逻辑准则将MV和DCT系数的检测结果结合起来，最终检测出视频中运动对象。实验结果表明，本文算法可准确地检测压缩视频中的运动对象，且检测结果具有较高的查全率和查准率。 |
| 关键词: H.264压缩域 运动矢量 DCT系数 特征累积 熵能 运动对象 |
| DOI： |
| 投稿日期： |
| 基金项目:国家863高技术研究发展计划资助项目(2009AA01Z328)；国家自然科学基金资助项目（60705013，60902091） |
|
| Associated features accumulation and analysis for motion object detection in compressed domain |
| LI Le1,2, ZHANG Maojun1, ZHANG Wenqi3, LI Yongle1 |
|
(1.College of Information Systems and Management, National University of Defense Technology, Changsha 410073, China;2.
2. Department of Information Engineering, Engineering University of Armed Police Force, Xi'an 710086, China;3. Satellite Launching Center in JiuQuan, Jiuquan 732750, China)
|
| Abstract: |
| A novel motion object detection method in H.264 compressed domain is proposed, based on multi-features accumulation and analysis, which can reflect the motion of object and change in the area of borders. Firstly, the MVs of macroblocks were accumulated and filtered. The Jacobian matrix was used to describe the relationship between parameters of global motion and MV in each macroblock, which makes the computation of global motion parameters easier than before. Then the motion areas were detected by the similarity of local motion and global motion. Secondly, the remarkable blocks of DCT energy were accumulated in temporal by the reliable motion around them. Then the border and texture area was found by the accumulations of DCT energy with the local self-adaptive threshold which selected by entropy. Finally, the results which had been detected by MV and DCT respectively were combined. The experiments shows that the method proposed can detect the motion object in compressed video accurately, and the result of can achieve good recall and precision simultaneously. |
| Keywords: H.264 compressed domain motion vectors DCT coefficients feature accumulation entropy energy motion object |
|
|
|
|
|
