基于时空方向主成分直方图的人体行为识别 下载: 1077次
Action Recognition Based on Histogram of Spatio-Temporal Oriented Principal Components
1 江南大学物联网工程学院, 江苏 无锡 214122
2 新疆大学电气工程学院, 新疆 乌鲁木齐 830047
图 & 表
图 1. 形成描述子HSTOPC的步骤
Fig. 1. Steps of HSTOPC descriptor generation
下载图片 查看原文
图 2. (a)高抛与(b)画圆动作的3D点云序列
Fig. 2. Action sequences of 3D point clouds of (a) high throw and (b) draw circle
下载图片 查看原文
图 3. HSTOPC的多尺度时域重叠分割
Fig. 3. Multi-scale temporal overlap segmentation method of HSTOPC
下载图片 查看原文
图 4. HOPCT+WHL算法在MSR Action3D数据集测试5/5实验结果的混淆矩阵
Fig. 4. Confusion matrix of the HOPCT+WHL algorithm by experiment setting of 5/5 on MSR action 3D dataset
下载图片 查看原文
表 1MSR Action3D数据库的识别率比较
Table1. Recognition rate comparison on MSR Action3D dataset
Method | Recognition rate /% |
---|
H3DF[16] | 89.45 | Pose Set[17] | 90.00 | HOG3D+LLC[18] | 90.90 | HKD[19] | 92.73 | DMM-LBP-DF[20] | 93.00 | HOPC[12] | 91.64 | HOPC+LHW | 93.77 | HOPC+LEE | 93.04 | HSTOPC+LHW | 94.51 | HSTOPC+LEE | 94.14 |
|
查看原文
表 2不同方法的实验结果比较
Table2. Comparison of experimental results with different methods
Method | Mean±STD | Max /% | Min /% |
---|
HOPC[12] | 87.06±10.63 | 96.55 | 66.67 | HOPC+LHW | 89.55±9.55 | 100 | 68.21 | HOPC+LEE | 89.71±9.80 | 100 | 68.42 | HSTOPC+LHW | 90.41±9.95 | 100 | 69.81 | HSTOPC+LEE | 90.56±9.05 | 100 | 71.93 |
|
查看原文
表 3ActionPairs3D数据库上的识别率比较
Table3. Accuracy comparison on MSR Action3D dataset
Method | Mean±STD | Max /% | Min /% | 5/5 |
---|
STPC[15] | - | - | - | 98.33 | HOPC[12] | 97.15±2.21 | 100 | 88.89 | 97.22 | HSTOPC+LEE | 98.18±2.75 | 100 | 90.56 | 98.33 |
|
查看原文
表 4UWA3D数据库的识别率比较
Table4. Accuracy Comparison with other description on UWA3D datasets
Method | Mean±STD | Max /% | Min /% |
---|
HON4D[20] | 79.28+2.68 | 88.89 | 70.14 | HDG[3] | 75.54±3.64 | 85.07 | 61.90 | HOPC[12] | 83.77±3.09 | 92.18 | 74.67 | HSTOPC+LHW | 85.11±3.21 | 94.04 | 75.67 |
|
查看原文
徐海洋, 孔军, 蒋敏, 昝宝锋. 基于时空方向主成分直方图的人体行为识别[J]. 激光与光电子学进展, 2018, 55(6): 061009. Haiyang Xu, Jun Kong, Min Jiang, Baofeng Zan. Action Recognition Based on Histogram of Spatio-Temporal Oriented Principal Components[J]. Laser & Optoelectronics Progress, 2018, 55(6): 061009.