Abstract: Systems and methods related to visual guidance for improving autonomous navigation with row following corrections in stereo camera systems are disclosed herein. A visual positioning system may implement a method for navigating a vehicle. The method may comprise: receiving navigation coordinates defining a route comprising a straight path; detecting detected lines that are parallel to the straight path based on images received from at least one camera on the vehicle; accessing ground truth data; deriving vehicle coordinates from the ground truth data; determining a centerline in vehicle coordinates based on the detected lines; determining corrected vehicle coordinates based on the centerline in vehicle coordinates and the vehicle coordinates from the ground truth data; and generating control signals to navigate the vehicle along the route using the corrected vehicle coordinates.

Abstract: Systems and methods related to soft obstacle detection are disclosed herein. The systems may include a stereo camera with two image sensors. An image from each of the two images sensors may be used to produce a geometric traversability map, which may be a depth map or a height map. An image from one of the image sensors may be used to produce a semantic traversability map based on semantically classifying portions of the image. The geometric traversability map and the semantic traversability map, each with element confidence scores, may be fused to create a fused traversability map. Specific elements of the images may be absent from the fused traversability map based on the characteristics of those elements. The fused traversability map may be a filtered 3D representation of the environment and may reduce false positives and false negatives in obstacle detection.

Abstract: This disclosure relates to the calibration and tuning of stereo cameras. A disclosed method includes measuring an operating temperature of a subject camera, applying the operating temperature to a subject camera function to produce an output, and tuning the subject camera at the operating temperature using the output. The subject camera functions can be derived using a calibration method. The calibration method can include obtaining reference camera measurement data at a set of temperatures and deriving a reference function for the reference camera using the data. The reference function uses temperature as an input. The calibration method can also include obtaining subject camera measurement data from the subject camera and deriving the subject camera function for the subject camera using the reference function for the reference camera and the subject camera measurement data.