Gm Cruise Holdings LLC

Robert Vets

Peter Megson

Zhanping Xu

Glenn D Sweeney

Kartheek Chandu

Adam Samuel Cadien

Ankit Rohatgi

Cornelis Jacob Lekkerkerker

Matt Cox

Mariana Rihl

Juan Sebastian Hurtado Jaramillo

Brandon S Seilhan

Alexander Lesnick

Systems and techniques are provided for identifying and filtering edge pixels from a 3D point cloud from a time-of-flight sensor. An example method includes receiving a first depth map that is based on an image frame captured by a time-of-flight (ToF) sensor, wherein the first depth map includes a plurality of measurements corresponding to a pixel array of the ToF sensor; generating a second depth map by shifting the plurality of measurements corresponding to the pixel array in at least one direction; comparing the first depth map with the second depth map to determine a measurement difference for each pixel in the pixel array; and identifying one or more edge pixels in the pixel array corresponding to at least one edge region in the image frame, wherein the measurement difference associated with the one or more edge pixels is greater than an edge threshold.

Detecting and filtering the edge pixels of 3D point clouds obtained from time-of-flight sensors

Aspects of the subject technology relate to systems, methods, and computer-readable media for identifying roll, pitch, and azimuth angles for an object based on surface normal vectors. Image data tracking a surface as an object moves relative to the surface can be accessed. A reference normal vector of the surface can be identified. A dynamic normal vector of the surface associated with the object moving over the surface can be identified from the image data. One or a combination of roll, pitch, and azimuth of the object with respect to the surface can be determined based on the reference normal vector and the dynamic normal vector.

Identifying stability of an object based on surface normal vectors

Aspects of the subject technology relate to systems, methods, and computer-readable media for performing ground segmentation through formation of a single super voxel. Image data tracking a surface as an object moves relative to the surface is accessed. Samples of the image data is grouped into different subsets to form a plurality of subsets of the image data. A normal vector to a corresponding plane defined by each subset of the plurality of subsets of the image data are identified. The samples of the image data are re-grouped into a single subset of the image data based on the corresponding normal vector for each subset of the plurality of subsets of the image data. A surface plane representation of the surface is identified based on the single subset of the image data.

Ground segmentation through super voxel

An autonomous vehicle having a LIDAR system mounted thereon or incorporated therein is described. The LIDAR system has N channels, with each channel being a light emitter/light detector pair. A computing system identifies M channels that are to be active during a scan of the LIDAR system, wherein M is less than N. The computing system transmits a command signal to the LIDAR system, and the LIDAR system performs a scan with the M channels being active (and N-M channels being inactive). The LIDAR system constructs a point cloud based upon the scan, and the computing system controls the autonomous vehicle based upon the point cloud.

Lidar system that generates a point cloud having multiple resolutions

Various aspects of the subject technology relate to a lidar validation system. The lidar validation system is configured to acquire one or more lidar point clouds from a lidar unit, wherein the lidar unit comprises one or more lasers, and wherein each point cloud is a representation of a scene according to a laser in the lidar unit, the scene comprising a fiducial target. The lidar validation system is configured to generate, based on points in the one or more point clouds, a target cloud that correspond to the fiducial target, perform a rigid body processing to minimize a sum of distances from each point in the target cloud to the fiducial target, and generate lidar validation measurements based on the distances from each point in the target cloud to the fiducial target.

Lidar sensor validation

LIDAR system that generates a point cloud having multiple resolutions

An autonomous vehicle having a lidar sensor system is described. A computing system is configured to determine that the lidar sensor system is to update a code that is included in light signals emitted by the lidar sensor system. The computing system transmits a command signal to the lidar sensor system, wherein the command signal causes the lidar sensor system to transition from emitting light signals with a first code therein to emitting light signals with a second code therein, wherein the first code is different from the second code.

Mitigating interference for LIDAR systems of autonomous vehicles

An autonomous vehicle having a LIDAR system that scans a field of view is described herein. With more specificity, a computing system of the autonomous vehicle defines a region of interest in the field of view for a scan of the field of view by the LIDAR system. The region of interest is a portion of the field of view. Based on the region of interest, the computing system transmits a control signal to the LIDAR system that causes the LIDAR system to emit first light pulses with a first intensity within the region of interest during the scan and second light pulses with a second intensity outside the region of interest during the scan. The first intensity is different from the second intensity to provide different ranges for distance measurements inside and outside the region of interest.

Growing community of inventors

Seattle, WA, United States of America

Robert Vets