The green forage harvester fills the trailer hoppers in real time by its mechanical arm when it is cutting and collecting green forage. Recently, the process of forage filling needs to identify the position of the trailer hoppers through artificial visual recognition, and then control the rotation of the mechanical arm to the right direction, which has the problems of low efficiency, high loss, and manpower consumption. Aiming at the demand of agricultural production intellectualization, a method of automatic recognition and location of trailer hopper of forage harvester based on threedimensional vision was proposed, which combined several advanced image processing methods with point cloud data processing technologies to realize edge recognition, spatial location of trailer hopper. Firstly, the concept of visual odometer was used to construct the relationship between the camera and the ground threedimensional coordinate system, and singular value decomposition (SVD) algorithm was used to calculate the pose transformation matrix, which was used to rotate and translate the three dimension (3D) point cloud under the camera coordinate, and threshold processing and dimension reduction were carried out based on the ground. Secondly, random sample consensus (RANSAC) algorithm was used to fit the edge of the hopper and locate the corners, so the relationship between the mechanical arm nozzle and the trailer hopper was determined. Finally, the result of localization was directly reflected on the pixel coordinates through coordinate transformation. The method proposed can accurately find the corners of the hopper and depict the area where the trailer hopper was located. The experimental results showed that the proposed method was in less computation, at the same time with high efficiency and accuracy, which satisfied the realtime and accuracy requirements in field operation.