An algorithm was proposed for recognizing apple fruit on the tree in order to eliminate the effects of different light and complex background. A kind of 3-D image of fruit tree was acquired with laser vision system. Image background was simplified by given scene knowledge. A double threshold detecting method was used for binarize image. Contour tracking technology based on chain code was used to record direction of target’s boundary. According to change rules of chain code information, relevant standards were established for extracting contour points of apple. Shape features of fruit were obtained by random circle method. The experimental results under various conditions showed that the algorithm was real-time. Its recognition rate reached to 93.75% while fruit area sheltered was less than 40%, and the whole effect was immune to light conditions and fruit surface color. 