RoboCup小型组比赛中多目标运动图像识别算法

A multi-target recognition algorithm for motion blur image in small-size RoboCup competition

由于隔行扫描摄像头采集的图像能产生运动模糊现象,因此采用隔行扫描摄像头的机器人足球比赛视觉子系统难以识别出场上的运动目标。在对航迹预测原理研究后,提出了多目标运动图像识别算法。该算法首先通过目标的当前位置预测目标下一周期的位置,在目标预测位置的区域内分别在奇数行和偶数行寻找目标像素,然后计算运动目标相邻两场之间水平方向和竖直方向的运动矢量,根据运动矢量采用插补方法进行去隔行处理,最后实现多目标同时识别。实验表明,采用这种方法能明显提高视觉系统对比赛中的运动目标的识别能力。

It is difficult for the visual subsystem of RoboCup robot soccer competition, which is realized by an interleaved camera, to identify the motion target on the field because of the motion fuzzy phenomenon. A new multi-target recognition algorithm for vision subsystem was presented based on the principle of track forecast. First, the target location of next frame was forecasted by current location and the target pixels on odd and even scan line were found in the region of forecasted target location. Then, the horizontal and vertical velocity vectors of the motion target were computed. Finally the motion interpolation was used to get high vertical resolution progressive frames from interlaced frames. Multi-target identity was realized at the same time. The result shows that visual system＇s identification ability of motion target can be improved by using this method.