Stereovision-based lateral offset measurement for vehicle navigation in cultivated stubble fields

In auto-guidance of agricultural vehicles in cultivated stubble fields, a precise measurement of the vehicle’s lateral offset could help achieve improved navigation accuracy. This paper presents an automated method for measuring a vehicle’s lateral offset for such an application. The basic concept was to use static ground features as references to detect the vehicle’s lateral offset. A stereo image-processing algorithm was developed which detected and tracked ground features captured in two consecutive field images, acquired using a vehicle-mounted stereo camera. These ground features were used as reference points to calculate the lateral offset. Field validation tests showed that this algorithm could provide accurate relative lateral offset measurements. Over a 10 m straight path, the deviation of measurements from ground truth was less than 50 mm. As this method was designed to measure relative offset, the field tests also revealed that it needed a realignment of the installation pose of the camera and the desired heading direction of the vehicle after a turn to ensure the measurement deviation less than 50 mm. Although the realignment algorithm was outside the scope of this research, the results proved the possibility of using an imaging sensor for the detection of the lateral offset of a vehicle to guide agricultural vehicles automatically in cultivated stubble fields without obvious referencing landmarks.