Abstract:

An algorithm for gesture detection and tracking in HCI (human-computer interaction) is designed to meet real-time and accuracy requirements. An innovative conception, which includes using distinguishing detection methods to detect hand-gesture for different conditions and using searching-box to decrease the searching zone, is proposed. The result shows that the detection rate can reach 97.36% while the missing rate lowers than 5%. It is robust to various external factors. However, it also meets the real-time, as the frame rate can reach 19.42. Compared with TLD, this algorithm has not only magnitude improvement in processing speed but also obvious advantages in accuracy.

Key words: object tracking, human-machine interaction, distinguishing detection, searching box

摘要：

为了降低识别复杂度, 提高识别效率, 实现手势的快速高效跟踪, 提出一种分情况检测思想和搜索框概念。首先对图像进行细检测, 得到目标的准确位置, 然后通过粗检测与跟踪相结合的方式进行目标跟踪, 并对跟踪结果进行修正和可信度判断。实验结果显示: 算法对图像手势的平均检测跟踪正确率可以达到
97.36%, 且保证平均漏检率在5%以下, 对各种外界因素具有较好的鲁棒性; 算法对视频图像的处理速度达19.42 帧/秒, 满足人机交互系统中的实时性需求; 与TLD 算法相比, 本算法在处理速度上有数量级的改善, 算法结果的正确率也有明显优势。

关键词: 目标跟踪, 人机交互, 分情况检测, 搜索框