Methods and systems for the fast estimation of three-dimensional bounding boxes and drivable surfaces using LIDAR point clouds

A system for an estimation of three-dimensional object bounding boxes in a three-dimensional point cloud, the system comprising:

• a processor comprising software executing coded instructions operable for:
• given a three-dimensional point cloud, identifying and removing a ground plane from the three-dimensional point cloud;
• clustering foreground points remaining after the ground plane is removed from the three-dimensional point cloud to generate one or more object clusters representing one or more objects;
• extracting one of a structural, textural, and density feature from each of the one or more object clusters representing the one or more objects using one or more of an Eigen value-based technique and an ensemble shape function-based technique;
• and
• fitting a three-dimensional bounding box around each of the one or more object clusters representing the one or more objects.

A method for identifying a similarly-shaped object in a feature-space associated with a three-dimensional point cloud, the method comprising:

• extracting a plurality of structural, textural, and/or density features from an object cluster representing an object using one or more of an Eigen value-based technique and an ensemble shape function-based technique;
• comparing the extracted plurality of structural, textural, and/or density features to a plurality of structural, textural, and/or density features extracted from one or more previously-detected objects using one or more of a combination classifier and deep learning model;
• and
• when a match is found, providing an object label to a three-dimensional bounding box as a classified pre-proposal for the object cluster representing the object.