激光传感器和机器视觉的复杂运动动作识别方法  

Complex motion recognition method based on laser sensor and machine vision

作　　者：黄斌[1] 王寅昊 HUANG Bin;WANG Yinhao(Guilin University of Technology,Guilin Guangxi 541004,China)

机构地区：[1]桂林理工大学,广西桂林541004

出　　处：《激光杂志》2023年第10期217-224,共8页Laser Journal

基　　金：广西壮族自治区科技计划项目(No.AK19125028)。

摘　　要：为解决因图像数据拟合、近似等情况引发的运动动作识别准确度低问题,研究激光传感器和机器视觉的复杂运动动作识别方法。通过激光传感器测量复杂运动加速度,通过相机采集复杂运动动作图像,且通过坐标系投影将其转换为二维坐标数据,结合最小二乘支持向量机和D-S证据理论,融合运动加速度数据以及图像坐标数据,将融合后的动作数据输入,基于LSTM-CNN的复杂运动动作识别模型中,通过CNN提取动作数据特征,将该特征输入LSTM中,利用LSTM深入提取动作数据时序特征,依据两种特征,LSTM的全连接层通过全连接层级联方法,获取复杂运动动作识别结果。实验结果表明:该方法可有效采集复杂运动动作图像,并实现复杂运动动作识别,且不同光照强度下的复杂运动动作识别精度较高,均不低于97%,且识别效率高,平均识别耗时为2.280 s。In order to solve the problem of low accuracy of motion recognition caused by image data fitting and approximation,a complex motion recognition method based on laser sensor and machine vision was studied.Complex motion acceleration is measured by laser sensor,complex motion image is collected by camera,and it is converted into two-dimensional coordinate data by coordinate projection.Combined with least square support vector machine and DS evidence theory,motion acceleration data and image coordinate data are fused,and the fused action data is input.In the complex motion recognition model based on LSTM-CNN,the feature of motion data is extracted by CNN,which is input into LSTM,and the time sequence feature of motion data is extracted by LSTM.According to the two features,the full-connection layer of LSTM obtains the recognition result of complex motion through the full-connection hierarchical method.The experimental results show that this method can effectively capture complex motion images and realize complex motion recognition,and the recognition accuracy of complex motion under different light intensity is high,which is not less than 97%,and the recognition efficiency is high,and the average recognition time is 2.280 seconds.

关 键 词：激光传感器 机器视觉 复杂运动动作 最小二乘 D-S证据理论 时序特征 

分 类 号：TN247[电子电信—物理电子学]