摘要
针对动态手势跟踪稳定性的不足和识别效率的问题,提出一种基于TLD和DTW的动态手势跟踪识别框架.首先利用基于Haar特征的静态手势分类器获得手势区域,然后使用TLD跟踪算法对获得的手势区域进行跟踪以获取手势轨迹,最后提取轨迹特征,使用改进的DTW算法进行识别.实验表明,该框架能够长时间稳定地跟踪手势区域,并能够在保证识别率的基础上显著提高识别效率.
For lack of stability in dynamic hand gesture tracking and low recognition efficiency, a hand gesture tracking and recognition framework is proposed based on TLD and DTW algorithm. First, the hand gesture area is got with a Haar features based classifier. Then the gesture trajectory is obtained by using the TLD tracking algorithm initialized by the hand area. Finally, features are extracted from the trajectory and the improved DTW algorithm is used to recognize the dynamic gesture. Experiments show that the framework can track the hand gesture stably for a long time and be able to improve the recognition speed greatly under the premise of ensuring high recognition accuracy.
出处
《计算机系统应用》
2015年第10期148-154,共7页
Computer Systems & Applications
关键词
动态时间规整
手势跟踪
手势识别
TLD
HAAR
dynamic time warping （DTW）
hand gesture tracking
hand gesture recognition
tracking-learning- detecting （TLD）
Haar
