人体骨骼数据容易受到环境中光线等干扰因素的影响,使得软件的姿态感知结果的均值平均精度(mAP)较高。因此,设计了一种基于光学跟踪和运动估计算法的新型VR头显实时姿态感知软件。以光学跟踪设备为核心组成测量系统,实时获得位姿测量数据,并经过误差校正后得到准确的姿态信息。借助于OpenPose网络检测出人体骨骼关键点,并应用引入提前终止策略的运动估计算法对实时姿态图像进行不断搜索,在忽略干扰因素的情况下找到最佳匹配点,以此来反映运动矢量分布位置,识别出当前人体姿态。最后,依托于多层感知机网络实现姿态分类,作为最终输出的实时姿态感知结果。测试结果表明,结合光学跟踪和运动估计算法的软件应用后,得出姿态感知结果的mAP值保持在0.95以上,满足了VR头显实时姿态感知要求。

Human bone data is susceptible to interference factors such as light in the environment,resulting in high mean average accuracy (mAP) of the software's posture perception results.Therefore,a novel VR head display real-time attitude perception software based on optical tracking and motion estimation algorithms is proposed.The measurement system is composed of optical tracking equipment as the core to obtain real-time pose measurement data,and accurate pose information is obtained after error correction.By utilizing the OpenPose network to detect key points of human bones,and applying a motion estimation algorithm that introduces an early termination strategy to continuously search for real-time pose images,the best matching point is found while ignoring interference factors,in order to reflect the distribution position of motion vectors and recognize the current human pose.Finally,relying on a multi-layer perceptron network to achieve attitude classification,as the final output of temporal attitude perception results.The test results show that after combining optical tracking and motion estimation algorithms,the mAP value of the attitude perception results remains above 0.95,meeting the real-time attitude perception requirements of VR head display.