摘要
为了解决现有行为检测系统中依赖惯性传感器、检测结果不够准确的问题,设计了基于人体骨架信息的行为检测系统;系统采用Jetson Nano人工智能计算设备作为主控模块,结合图像采集模块、显示模块和以Atmega328单片机为主的报警模块构成;系统利用图像采集模块采集行为视频信息,通过主控模块中的行为检测器对视频中人体行为进行检测,报警模块通过串口接收检测结果并对危险行为进行预警;同时,利用人体骨架的关节空间运动幅度、肢体关联差异,建立了关节帧间位移矢量和骨骼夹角变化的关节行为模型,再借助长短时记忆网络提取行为特征,并训练实时行为检测器;经实验测试,该系统能够有效检测常见的人体行为并对危险行为类别进行报警提示。
In order to solve the problems that the existing action detection system relies on inertial sensors and the detection results are not accurate enough,an action detection system based on human skeleton information is designed.The system uses Jetson Nano artificial intelligence computing equipment as the main control module,the auxiliary modules consist of the image acquisition module,display module and alarm module,alarm module based on Atmega328P microcontroller.The system uses the image acquisition module to collect the action video information,detects human action in the video through the action detector in the main control module,and the alarm module receives the detection results through the serial port and gives early warning of dangerous actions.At the same time,using the joint space motion range of the human skeleton and the difference of limb associations,a joint action model for the displacement vector between joint frames and changes in the bone angle is established,and then the long short term memory(LSTM)network is used to extract the action features and train the real time action detector.After the experimental test,the system can effectively detect common human action and give alarms to dangerous action categories.
作者
雷振轩
LEI Zhenxuan(College of Information and Control Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China)
出处
《计算机测量与控制》
2023年第1期30-35,44,共7页
Computer Measurement &Control
基金
陕西省重点研发计划社会发展领域项(2022SF-242)。
