Transportation for harvest and plant protection activities inside greenhouse calls for a lot of labor. The available transporting equipment has limited function which is not very flexible. While the widely used railway transportation vehicles inside greenhouse in many developed countries are hard to introduce to China due to variable construction styles and nonstandard layouts, some automatic following system is urgently in need. Thus a kind of autonomous transportation system for greenhouse harvesting activity was presented based on Kinect sensor developed by Microsoft Company. The Kinect sensor could acquire the depth data of each pixel on the picture, from which the skeleton of a workman could be analyzed by several image processing algorithms. Direction and heading angle was determined by the real-time coordinate of the center of gravity of the skeleton. At the same time, the 3D coordinate of the workman was kept and the tracking path was thus formed. To avoid dramatic turning, adaptive function was applied to optimize the path firstly. And then pure pursuit method was introduced to complete the tracking stably and precisely. The look-ahead distance was timely decided by fuzzy controller, with which the vehicle would track more flexibly according to different lookhead distances. The result showed that the system was able to turn freely and precisely while avoiding drastic turning. The largest lateral deviation of straight line tracking was smaller than 10.0cm, and the largest depth deviation was 5.5cm. The working function had nothing to do with the light condition, which met the demands of harvesting and goods transportation in greenhouse in different seasons.