To solve the key prerequisite of greenhouse carrier vehicle navigation, a new navigation method was proposed based on fixed direction camera. A pan-tilt camera was constructed to collimate the vision line to corridor’s center line with geomagnetic sensor and digital servo. The angle error was detected by geomagnetic sensor and the error was compensated by servo. The drawback of Hough transform was avoided by using new method of picture processing. The vision system ran in Raspberry Pi based on OpenCV in Linux. It distinguished corridor and crop by value of hue instead of Excess—Green method and the end point of corridor was got by area of screening. By using the Ackerman principle, the vehicle’s position and attitude angle can be received. The dynamic response was got by modeling the equation of vehicle and simulating the movement of vehicle at different vehicle speeds. The experiment was done in real greenhouse which was divided into two parts, including the static experiment and dynamic experiment. The error of navigation information was obtained by processing the picture in fixed platform and a control experiment with Hough transform. The vehicle can drive automatically at different speeds in greenhouse. The results showed that the average time cost of this method was less than 240ms/f, the average detection error of the navigation line was 2.4cm, and the long-focus camera can reduce the error by 25.7%, which indicated that the method was available to get navigation line with appropriate accuracy and instantaneity.