人体姿态识别算法在视觉人机交互中的应用

Application of human body posture recognition algorithm in visual Human-Computer Interaction

摘要:

针对当下视觉人机交互应用中所面临的识别过程复杂，交互效率低等一系列问题，提出了一种基于人体构造的姿态识别算法。利用该算法和软硬协同的设计思想，在FPGA平台上实现了一套运用视觉感知进行交互的机器人控制系统。图像采集模块获取双摄像头的画面，图像预处理模块完成降噪和冗余帧的去除，关节识别模块实现对人体皮肤和关节处颜色标记的检测；得到的关节位置信息送入软核，运用空间向量模型识别出肢体动作。最终将识别结果编码后通过蓝牙传输给远端机器人。实际测试表明，该算法能快速准确地识别出绝大部分基本的肢体动作，机器人能实时跟随人体简单运动。

Abstract:

Aiming at a series of problems in the current visual human-computer interaction application, such as complex recognition process and low interaction efficiency, This paper proposed a pose recognition algorithm based on human body structure. Using the algorithm and the design idea of soft and hard coordination, this paper realized a set of robot control system using visual perception to interact on the FPGA platform. Image acquisition module got pictures of the dual camera, the image preprocessing module completed the noise reduction and the removal of the redundant frame, and the joint recognition module realized the detection of the color mark on the human skin and joint; Soft nuclear the obtained joint position information, using spatial vector models to identify limb movements. The soft nuclear encoded the recognition result and transmitted to the remote robot via Bluetooth. The actual test shows that the algorithm can quickly and accurately identify most of the basic limb movements, and the robot can follow the simple movement of the human body in real time.