Abstract: In order to enhance the localization accuracy of harvesting robots’ vision system, a ranging method based on binocular stereo vision technology was presented, which used a combined matching method and a calibration model for depth values. After centroid-based matching, rough parallax can be acquired to set the parallax range during area-based matching. In this way, calculated amount and error matching probability can be reduced. Then the depth map of tomato region can be obtained through triangulation ranging principle after area-based matching. Depth mean value in tomato region is regarded as the distance between tomato and camera. Experimental results showed that ranging errors and distances were linear relative, so linear regression models were set up to calibrate the ranging results after regression analysis. The ranging error was about -7~5 mm when the distance was less than 650 mm. And the ranging error was about ±10 mm when the distance was less than 1 050 mm. It can meet the need of vision system of tomato harvesting robot in most harvesting environments.