The recognition of litchi clusters and the calculation of picking point in a nocturnal natural environment

In a natural environment, the recognition of ripe litchi and calculation of the picking point are always difficult problems for a picking robot. In this study, a visual system for litchi image acquisition is built and a method of nocturnal litchi recognition and a calculation of picking point is proposed. For comparison, images of the same cluster of litchis are captured during the day in a natural environment and during the night using artificial illumination. By analysing colour features of the same litchi image in different colour models, the YIQ colour model is proved to be the model with the best practicability for nocturnal litchi recognition. In this proposed method, the background of the nocturnal image, instead of the litchi fruit and stem, is first removed using an improved fuzzy clustering method (FCM) combining this analysis approach with a one-dimensional random signal histogram. The fruit is then segmented from the stem base using the Otsu algorithm. The Harris corner was used for picking point detection. The change rates of the horizontal and vertical positions between corner points are analysed to identify the picking point. The experiments show that nocturnal litchi recognition accuracy is 93.75% with an average recognition time of 0.516 s. At different depth distances, the highest accuracy for the picking point calculation is 97.5%, while the lowest is 87.5%. The results show the accuracy and feasibility of this method for litchi recognition and picking point calculation during the night. This research provides technical support of visual localisation technology for litchi-picking robots.