苹果采摘机器人目标果实的快速跟踪识别

Fast tracking and recognition of apple harvesting robot for target fruit

为了实现苹果采摘机器人对果实的精确快速采摘,文章在压缩跟踪算法的基础上提出一种结合SURF特征的跟踪方法。该方法首先使用k-means聚类算法分割出初始帧中目标苹果,然后根据目标苹果的位置和尺度信息,生成随机稀疏矩阵并训练生成初始的贝叶斯分类器,最后通过SURF特征检测对下一帧中的目标苹果进行跟踪匹配,同时更新分类器参数从而实现目标苹果果实的跟踪识别。通过实验比较,此方法在目标苹果尺度发生变化和受到外界遮挡的情况下影响较小,满足采摘要求的准确性与实时性。

In order to realize the accurate and quick picking of apple harvesting robot,this paper proposes a tracking method based on the SURF feature based on the compression tracking algorithm.Firstly,the k-means clustering algorithm is used to segment the target apples in the initial frame,and then the stochastic sparse matrix is generated according to the position and scale information of the target apple. The initial Bayesian classifier is trained and finally passed the SURF feature in the target apple to track the match,while updating the classifier parameters in order to achieve the target apple fruit tracking identification. Through the experimental comparison,this method has little effect on the change of the target apple scale and the outside occlusion,which satisfies the accuracy and real-time of the picking requirement.