Abstract: Vision system is one of key parts of agricultural harvesting robots, which provides the fruit's position information for navigating the manipulator. Considering its applications in agriculture and the pineapple is big enough for recognition, this study presents a low-cost binocular vision platform for pineapple harvesting robots, which consists of low-cost CMOS (Complementary Metal Oxide Semiconductor) image sensors, a tripod, a binocular pan, a PC and software system; The calibration model and the calibration software were developed based on C++ and OpenCV version 1.1, and Matlab calibration toolbox separately; The Zhang's algorithm was employed during the calibration. By experiment, the suitable calibration method for the constructed platform was selected. Based on the low-cost vision platform and developed pineapple recognition algorithms, 3D position calculation experiments for pineapples were conducted in a pineapple field of Zhanjiang. The results showed that the depth errors were less than 6-8 cm when the depth distance was around 1 m, and the errors were less than 2-3 cm after correcting the whole system. The low-cost platform performed well and its feasibility was proved. This study can provide a reference for the development of pineapple harvesting robots.