Chinese Journal of Liquid Crystals and Displays, Volume. 37, Issue 4, 530(2022)
Campus violence action recognition based on lightweight graph convolution network
Figures & Tables(12)
Fig. 7. Variation diagram of model training loss and test accuracy
Table 1. Six typical action recognition results
View tableView in ArticleTable 1. Six typical action recognition results
动作类别 样本个数/个 识别准确率/% 最大识别速度/fps 拳打 200 94.5 20.6 脚踢 200 97.0 20.6 倒地 200 98.5 20.6 推搡 200 95.0 20.6 打耳光 200 94.5 20.6 跪地 200 95.5 20.6
Table 2. Action recognition results of personnel with different body types
View tableView in ArticleTable 2. Action recognition results of personnel with different body types
测试集人员 身高/cm 体重/kg 肩宽/cm 平均识别
准确率/%
1号 173 76 45 83.5 2号 179 67 41 68.7 3号 154 46 35 72.4 4号 168 59 37 85.6 5号 176 80 48 85.0 6号 163 103 47 64.5
Table 3. Comparison results of different recognition methods
View tableView in ArticleTable 3. Comparison results of different recognition methods
方法 参数量/MB 平均识别准确率/% 最大识别速度/fps Slow Fusion CNN［ 14 ］17.03 65.4 15.5 VA-CNN［ 15 ］24.03 82.8 9.2 ST-GCN［ 7 ］3.10 85.6 12.7 Ours（无注意力模块） 1.25 86.8 18.7 Ours（含注意力模块） 0.78 89.7 19.3
Table 4. Six typical action recognition results
View tableView in ArticleTable 4. Six typical action recognition results
动作类别 识别准确率/% 拳打 85.2 脚踢 90.7 行走 92.4 下蹲 88.2 跳远 87.6 推 91.5
Tools
Get Citation
Copy Citation Text
Qi LI, Yao-hui DENG, Jiao WANG. Campus violence action recognition based on lightweight graph convolution network[J]. Chinese Journal of Liquid Crystals and Displays, 2022, 37(4): 530
Paper Information
Category:
Received: Aug. 31, 2021
Accepted: --
Published Online: Jun. 20, 2022
The Author Email: Yao-hui DENG (173743077@qq.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20