基于骨骼信息下的手势识别研究

GESTURE RECOGNITION BASED ON SKELETON INFORMATION

在人体行为动作中,基于手势动作的识别研究具有便捷、直观、互动性强、表达信息丰富等特点而成为众多研究者对人体行为分析的首选。传统的动作识别一般是建立在二维彩色图像上进行研究,而二维RGB图像容易受到背景扰动、光线、环境等因素对人体目标检测的干扰。对微软Kinect体感设备进行研究,利用Kinect捕获的深度图像,将其转换成骨骼图并提取骨骼关节点空间坐标信息,设计一个人体手势识别系统。借助动态时间规整算法将测试模板与训练样本进行相似度的匹配计算,达到姿势识别的目的。在Unity 3D里进行虚拟模型同步演示,实现了对人体动作手势指令的识别和对模型的控制。

In human behavior, the recognition research based on gesture has many characteristics such as convenience, intuition, strong interaction, and rich expression information, so it has become the first choice for many researchers to analyze human behavior. Traditional motion recognition is generally based on two-dimensional color image, and two-dimensional RGB images are easily interfered by background disturbance, light, environment and other factors when human are detected. In the paper, we studied Microsoft Kinect somatosensory device which was used to capture depth image, converted the depth image into bone image to get spatial coordinate information of bone joint point, and designed the human body gesture recognition system. The dynamic time warping algorithm was used to calculate the similarity between the test samples with the training samples, so as to achieve the purpose of gesture recognition. The virtual model synchronization demonstration was implemented in Unity 3D, which realized the recognition of human gesture instructions and the control of model.