Michael J Harrell

US Patent:

20150379371, Dec 31, 2015

Filed:

Jun 30, 2014

Appl. No.:

14/320053

Inventors:

- Redmond WA, US
Aaron Rogan - Westminster CO, US
Michael Harrell - Denver CO, US
Bradford R. Clark - Broomfield CO, US

International Classification:

G06K 9/52
G06K 9/62
G06K 9/46

Abstract:

Two-dimensional and three-dimensional data of a physical scene are combined and analyzed together to identify physical objects physically present in the physical scene. Image features obtained from the two-dimensional data and geometric features obtained from the three-dimensional data are combined with one another such that corresponding image features are associated with corresponding geometric features. Automated object detection mechanisms are directed to the combination of image and geographic features and consider them together in identifying physical objects from the physical scene. Such automated object detection mechanisms utilize machine learning such as selecting and tuning multiple classifiers, with each classifier identifying potential objects based on a specific set of image and geographic features, and further identifying and adjusting weighting factors to be applied to the results of such classifiers, with the weighted combination of the output of the multiple classifiers providing the resulting object identification.

US Patent:

20150362587, Dec 17, 2015

Filed:

Jun 17, 2014

Appl. No.:

14/306755

Inventors:

- Redmond WA, US
Benjamin James Kadlec - Boulder CA, US
Michael Riley Harrell - Denver CO, US

International Classification:

G01S 7/497
G01S 17/89

Abstract:

Lidar scanning is used in a variety of scenarios to detect the locations, sizes, shapes, and/or orientations of a variety of objects. The accuracy of such scanning techniques is dependent upon the calibration of the orientation of the lidar sensor, because small discrepancies between a presumed orientation and an actual orientation may result in significant differences in the detected properties of various objects. Such errors are often avoided by calibrating the lidar sensor before use for scanning, and/or registering the lidar data set, but lidar sensors in the field may still become miscalibrated and may generate inaccurate data. Presented herein are techniques for identifying, verifying, and/or correcting for lidar calibration by projecting a lidar pattern on a surface of the environment, and detecting changes in detected geometry from one or more locations. Comparing detected angles with predicted angles according to a predicted calibration enables the detection of calibration differences.