基于改进Hu矩算法的Kinect手势识别  

Kinect Gesture Recognition Based on Improved Hu Moment Algorithm

作　　者：蒲兴成[1] 王涛[2] 张毅[2] 

机构地区：[1]重庆邮电大学理学院,重庆400065 [2]重庆邮电大学自动化学院,重庆400065

出　　处：《计算机工程》2016年第7期165-172,180,共9页Computer Engineering

基　　金：科技部国际合作基金资助项目(2010DFA12160);重庆市教委科学技术研究基金资助项目(KJ1400432)

摘　　要：针对Kinect深度信息下静态手势识别鲁棒性差及识别率低的问题,在改进Hu矩算法的基础上提出新的手势识别方法。构造一种新的Hu不变矩,采用尺度归一法消除尺度变化对Hu矩的影响,使静态手势在离散状态下保持手势的比例、平移以及旋转不变性。通过微软新型Kinect传感器和Open NI获取深度图像,对其进行去噪等预处理,基于灰度直方图进行手势分割,提取手指个数特征,并利用改进的Hu矩不变矩算法识别静态手势。实验结果表明,该方法在光照变化、复杂背景等干扰下具有强鲁棒性和高识别率。In order to solve the problem of poor robustness and low recognition rate in static gesture recognition based on the Kinect depth information,this paper presents a new gesture recognition method based on the improved Hu moment algorithm. It constructs a new invariant Hu moment by using the scale normalization method to eliminate the influence of scale on Hu moment, which can keep the static gestures＇ invariability with shifting, scaling and rotation under the discrete states. Moreover,the Kinect sensor and Open NI are used to get the depth information of images. The obtained depth information is pre-treated, such as the noise elimination. Then the histogram method is used for gesture segmentation and feature extraction of the number of fingers. Finally, it uses the improved Hu moment algorithm to recognize static gestures. Experimental results show that the improved algorithm has stronger robustness and higher recognition rate even under complicated backgrounds.

关 键 词：传感器 静态手势 深度信息 特征提取 HU矩 手势识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]