Tomato targets extraction and matching based on computer vision

Tomato targets extraction and matching is the basis for tomato location and picking of tomato harvesting robot. It is a challenge to match the targets extracted from images captured by different cameras, when the number of the tomatoes is more than three, especially with the presence of clutter and occlusion. A novel tomato targets extraction and matching algorithm is proposed in this paper to solve this problem. Firstly, the image segmentation algorithm is developed to segment the targets from the background, for the tomato images captured based on color analysis. Secondly, a method for slip touching tomatoes separation based on gray-scale local maxima is used to estimate the number and radius of the tomatoes based on the extracted tomato area. Then, RCR (Random Circle Ring) method is applied to extract the center and radius of the tomatoes. The last but not the least, tomato targets extracted from the image captured by the first camera are matched with the ones in the image captured by the second camera through applying SURF (speeded up robust features) algorithm. Experimental results showed that the approach proposed in this paper solved the extraction and matching of tomato targets with the presence of clutter and occlusion to some extent, while achieving obvious validity and accuracy.