基于Camshift反馈码本模型的手抓取物跟踪方法

Tracking Method of Hand Grasping Objects Based on Feedback from Camshift to Codebook Model

针对Camshift算法在复杂背景下无法自动跟踪目标的问题,提出一种基于Camshift反馈码本模型的运动目标检测和跟踪算法。该算法首先利用码本模型检测前景目标,然后采用Camshift在颜色概率分布图中跟踪前景区域中的目标,通过窗口尺寸比较和直方图相关性判断来解决自动跟踪,通过窗口位置预测和尺寸扩大来改进下一帧Camshift算法的输入搜索窗口,同时并集操作多个目标处理后的矩形窗口,并将其反馈为下一帧码本模型的图像检测区域。最后将该算法应用于手和目标物的抓取状态判断上,具体过程是在静态背景下利用两个摄像头采集到的图像进行手和目标物的检测和跟踪,通过矩形相交性判断抓取次数,以验证跟踪算法的有效性。实验结果表明,通过信息反馈减小了目标检测和跟踪的搜索区域,提高了算法的实时性,在单摄像头下可提高处理帧频130%。

In order to solve problem that Camshift algorithm can＇t automatically track targets in complex background, an algorithm to detect and track moving objects based on feedback from Camshift to codebook model was proposed. At first, the algorithm detects foreground objects by codebook model and tracks objects in the foreground area by Camshift using color probability distribution. Automatic tracking is achieved by comparison of window sizes and judgment of histogram correlation, and an input search window of the next frame is improved by location prediction and size expansion of window. At the same time,a union of multi-processed rectangular windows is as a feedback to the next frame of the image detection region for codebook model. Finally, the algorithm is applied for determining a grasping status between a hand and objects. The specific process is using the captured images by two cameras to detect and track hand and objects in a static background and then counting the number of grasp by rectangle intersection to verify the tracking algorithm. Since the information feedback reduces the search region of detection and tracking, this algorithm gets a better real-time performance, and some experimental results show that the frame processing rate can increase 130% in a single camera.