基于傅立叶描述子和HMM的手势识别

Hand Gesture Recognition Based On Fourier Descriptor And HMM

针对家庭服务机器人平台中人机交互的问题,提出基于视觉的手势识别作为人与机器人交互的方式,研究利用傅立叶描述子对手势形状进行描述,并结合支持向量机和隐马尔可夫模型分别对静态手势和动态手势进行分类,实现了静态手势和动态手势的识别。该系统基于新型传感器Kinect,在图像分割阶段结合图像深度信息,可以有效的将手势区域提取出来,在一定范围内具有较强的鲁棒性,特征提取阶段基于傅立叶描述子,使手势识别具有旋转、缩放、平移不变性。针对七种常见静态手势和四种动态手势进行测试,平均识别率分别达到98.8%和96.7%,实验结果表明该系统具有较高的准确度。

In order to solve Human and Robot interaction problem on the platform of home service robot, an interaction method for robot and human based on hand gesture recognition is presented. It comes up with a method to describe hand posture by Fourier Descriptor, combined with Supported Vector Machine and Hidden Markov Mode/to recognize static hand posture and dynamic hand gesture. A novel camera sensor （Kinect） is used. Hand region is extracted with range image information and skin colour model in the phase of image segmentation. Fourier Descriptor that is independent of rotation, scaling and translation, is utilized for feature extraction. The experi- ment is conducted on seven common static hand postures and four dynamic hand gestures, and each of the average recognition rates is 98.8 % and 96. 7 %. The result demonstrates that the system has high recognition rate and strong robustness.