To automatically recognize the geometric parameters of fresh tea leaves with different orientations, constrained Delaunay triangulation was introduced to triangulate the area of fresh tea leaf after operations, which made the contour points distribute evenly and sparsely. Then, with the character of end triangle, adjacent triangle and connect triangle, median axis of tea leaves and tea stalks were calculated, thus the orientation of a fresh tea leaf was determined and the tea leaf was labeled. At the same time, the length, width, and area of each tea leaf was calculated, as well as distance between petioles. Based on above results, a geometric feature index concerning the oldness of fresh bulk tea was proposed, and a possible method to finely divide the grade of fresh tea was also discussed. Totally 150 images with 174 pieces of fresh tea leaves were used to validate the algorithm, and a recognition rate of 94.2% was obtained, and the average time to treat a single fresh tea leaf was 74.7ms.