作者：何赟泽1, 周辉1, 吴兴辉1, 任丹彤1, 丁美有1, 程亮2,3
作者单位：1. 湖南大学电气与信息工程学院, 湖南 长沙 410006;
2. 江苏海洋大学海洋工程学院, 江苏 连云港 222005;
3. 珠海云洲智能科技有限公司, 广东 珠海 519085
关键词：水域环境;行为识别;水域人员动作行为数据集;SlowFast算法改进
摘要：
为保障水域人员生命安全，针对水域环境人员不安全行为识别问题，提出一种基于Faster-RCNN+改进SlowFast的水域人员行为识别网络。主要工作包括以下几点：第一：根据采集的95段视频制作水域人员动作行为数据集，其中共有161687张快通道帧，5543张关键帧，共计标注框9173个。第二：使用Faster-RCNN+SlowFast模块化设计对水域人员动作行为数据集进行目标定位及行为分析，其实验结果表现良好，可以识别水域人员位置及行为。第三：提出一种与时域压缩策略相结合方式的改进SlowFast网络，结果表明改进SlowFast网络模型识别准确率提高5.0%，推理速度得到约1.14倍提升，实验证明可有效帮助识别水域人员的行为识别问题。
Unsafe behavior recognition algorithm and application for water personnel
HE Yunze, ZHOU Hui, WU Xinghui, REN Dantong, DING Meiyou, CHENG Liang
1. College of Electrical and Information Engineering, Hunan University, Changsha 410006, China;
2. School of Ocean Engineering, Jiangsu Ocean University, Lianyungang 222005, China;
3. Zhuhai Yunzhou IntelligentTechnology Co., Ltd., Zhuhai 519085, China
Abstract: In order to ensure the life safety of water personnel, a behavior recognition network based on Faster-RCNN+SlowFast improved was proposed for unsafe behavior recognition of water personnel. The main work includes the following points: First, according to the 95 videos collected, the dataset of water area personnel's action behavior is constructed, in which there are 161687 fast channel frames, 5543 key frames and a total of 9173 labeling frames. Secondly, the Faster-RCNN+SlowFast modular design is used to conduct target location and behavior analysis on the dataset of personnel's action behavior in water area. The experimental results show good performance and can identify the location and behavior of personnel in water area. Thirdly, an improved SlowFast network combined with time-domain compression strategy is proposed. The results show that the recognition accuracy of the improved SlowFast network model increases by 5.0%, and the inference speed increases by about 1.14 times. The experiment proves that the improved network model can effectively help to identify the behavior recognition problms of people in water areas.
Keywords: water environment;behavior recognition;water personnel action behavior dataset;improvement of SlowFast
2023, 49(9):104-110 收稿日期: 2022-6-28;收到修改稿日期: 2022-9-2
基金项目:
作者简介: 何赟泽（1983-），男，山西祁县人，教授，主要研究方向为嵌入式人工智能与边缘计算、红外热成像与机器视觉。
参考文献
[1] 杨刚, 张宇姝, 宋震. 人体动作识别与评价——区别、联系及研究进展[J]. 计算机科学与探索, 2022, 16(5): 991-1007.
[2] 杜思远, 李杰, 郑涛, 等. 基于三子样更新的EKF低成本MEMS姿态估计算法[J]. 中国测试, 2018, 44(3): 38-43.
[3] 郭海艳, 程亮, 杨春利, 等. 面向水面无人艇的目标检测与船舶分类系统研究[J]. 中国测试, 2023, 49(6): 114-121.
[4] 黄晴晴, 周风余, 刘美珍. 基于视频的人体动作识别算法综述[J]. 计算机应用研究, 2020, 37(11): 3213-3219.
[5] 乔羽. 基于Mask R-CNN泳池中溺水行为检测系统的设计与实现[D]. 青岛: 青岛大学, 2019.
[6] FEICHTENHOFER C, FAN H, MALIK J, et al. Slowfast networks for video recognition[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019.
[7] NIU J, YANG R, GUAN W, et al. Spatial-temporal graph convolutional networks for action recognition with adjacency matrix generation network[C]//2nd International Conference on Electronics, Communications and Information Technology (CECIT), 2021.
[8] HUYNH T T, HUA C H, TU N A, et al. Learning 3D spatiotemporal gait feature by convolutional network for person identification[J]. Neurocomputing, 2020, 397(3): 192-202.
[9] TAN D, HE C, WANG Y. Action recognition model based on feature interaction[C]//Proceedings of the 11th International Conference on Computer Engineering and Networks, 2021.
[10] ULLAH A, AHMAD J, MUHAMMAD K, et al. Action recognition in video sequences using deep bi-directional LSTM with CNN features[J]. IEEE access, 2017, 6(99): 1155-1166.
[11] 杨清山, 穆太江. 采用蒸馏训练的时空图卷积动作识别融合模型[J]. 中国图象图形学报, 2022, 27(4): 1290-1301.
[12] SIMONYAN K, ZISSERMAN A. Two-stream convolutional networks for action recognition in videos[J]. Advances in neural information processing systems, 2014, 27(1): 568-576.
[13] WU H, SONG C, YUE S, et al. Dynamic video mix-up for cross-domain action recognition[J]. Neurocomputing, 2022, 471(6): 358-368.
[14] GU C, SUN C, ROSS D A, et al. Ava: A video dataset of spatio-temporally localized atomic visual actions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018.
[15] 郭雨青, 曾庆军, 夏楠, 等. 图像增强水下自主机器人目标识别研究[J]. 中国测试, 2021, 47(11): 47-52.
[16] 彭伟康, 陈爱军, 吴东明, 等. 基于改进Faster R-CNN的水准泡缺陷检测方法[J]. 中国测试, 2021, 47(7): 6-12.
