自然场景下果实目标的识别和定位

The recognizing and locating method for fruit objects under nature scenes

通过分析目前果蔬图像中识别果蔬对象存在的问题,研究了果蔬采摘机器人视觉中的两大难题:识别果蔬和定位果蔬.在此大背景下,研究提出了一种结合多种颜色特征和纹理特征进行分割的方法,有效解决果蔬目标和背景颜色差异较小时的果蔬对象识别问题.同时,提出一种全新的理念解决果实被部分遮挡时中心点和采摘点确定问题.另外,为了克服已有的类球形果蔬的定位方法不能适用于偏斜下垂生长的果蔬定位、不能有效定位叠加的多个果蔬的不足,利用几何校正方法,研究提供一种能够适应于偏斜下垂生长的果蔬定位的方法,能有效定位叠加的多个果蔬.经实验证明,效果良好.

This paper discussed about the most difficult problems in the vision of harvesting robot, including recognizing and locating problem existed in the recognition of the fruit object in fruit images. A new segmentation method combined with color features and texture features is presented, in which the recognition problem with little difference between with the target and the background is solved. It also put forward a novel conception to resolve the locating center and abscission point under the condition that the fruit or vegetables are partially sheltered. Meanwhile, in order to overcome the demerits of the going locating methods which unfits for the declining fruit, a new locating method based on the technique of geometry correction fitting for the decli- ning fruits and locating the abscission points under the condition that fruits are overlapped by each other is presented. It is proved by experiments that it will work well under the nature scenes.