A new algorithm for greenhouse corridor edge detection with RGB-D data

This paper presents a new algorithm of corridor edge detection in greenhouse scenes with RGB-D sensor. The algorithm tries to tackle the detection problem considering corridor's 3 features, including 3D position, dominant color and binary edges. Firstly, a height interval will be used to collect 3D points for generating a corridor candidate set. Secondly, dominant color abstracting will help to delete outliers with similar height but different colors. And connected region labeling will help to get rid of outliers with similar height and color but different locations. After that dominant color abstracting and connected region labeling will be repeated on the whole image set, and the resulted region will have different edges from the one generated from corridor candidate set. Then the common edges will be taken as root edges. Thirdly, a directional growth strategy will finally mark out the whole corridor edges on canny edges map, with root edges as its start. Experiments are carried out on real RGB-D data collected in greenhouse, and the effectiveness of the new algorithm is verified.