Systems and methods for detecting objects based on lidar data
AssigneesDENSO International America, Inc., Denso Corporation
InventorMinglei Huang
Life of this patent
Estimating the term…
Abstract
A method includes generating a plurality of lidar inputs based on the lidar data and performing, for each lidar input from among the plurality of lidar inputs, a convolutional neural network (CNN) routine to: generate a plurality of bounding boxes based on the lidar inputs, determine one or more dimensional characteristics associated with each of the plurality of bounding boxes, and generate a confidence score associated with each of the plurality of bounding boxes. The method includes selecting one or more target bounding boxes from among the plurality of bounding boxes based on the confidence score associated with each of the plurality of bounding boxes and identifying the one or more objects based on the one or more target bounding boxes.
Claims10
A method for detecting one or more objects based on lidar data obtained from a lidar sensor of a vehicle, the method comprising:
- generating a plurality of lidar inputs based on the lidar data;
- performing, for each lidar input from among the plurality of lidar inputs, a convolutional neural network (CNN) routine to:
- generate a plurality of bounding boxes based on the lidar inputs;
- determine one or more dimensional characteristics associated with each of the plurality of bounding boxes;
- and
- generate a confidence score associated with each of the plurality of bounding boxes, wherein generating the confidence score includes performing an Intersection over Union (IoU) routine to determine an overlap between each of the plurality of bounding boxes and one or more ground truth bounding boxes associated with the one or more objects, and generating the confidence score based on the overlap,
- wherein the overlap indicates a matching between the one or more dimensional characteristics of the bounding box and the ground truth bounding box, the one or more dimensional characteristics including one or more of a center point, a length, a width, and a yaw angle;
- selecting one or more target bounding boxes from among the plurality of bounding boxes based on the confidence score associated with each of the plurality of bounding boxes by iteratively performing a non-maximum suppression (NMS) routine based on the confidence score associated with each of the plurality of bounding boxes until the one or more target bounding boxes do not overlap;
- and
- identifying the one or more objects based on the selected one or more target bounding boxes.
A system for detecting one or more objects based on lidar data obtained from a lidar sensor of a vehicle, the system comprising:
- one or more processors and one or more non-transitory computer-readable mediums storing instructions that are executable by the one or more processors, wherein the instructions comprise:
- generating a plurality of lidar inputs based on the lidar data;
- performing, for each lidar input from among the plurality of lidar inputs, a convolutional neural network (CNN) routine to:
- generate a plurality of bounding boxes based on the lidar inputs,
- determine one or more dimensional characteristics associated with each of the plurality of bounding boxes, and
- generate a confidence score associated with each of the plurality of bounding boxes, wherein generating the confidence score includes performing an Intersection over Union (IoU) routine to determine an overlap between each of the plurality of bounding boxes and one or more ground truth bounding boxes associated with the one or more objects, and generating the confidence score based on the overlap,
- wherein the overlap indicates a matching between the one or more dimensional characteristics of the bounding box and the ground truth bounding box, the one or more dimensional characteristics including one or more of a center point, a length, a width, and a yaw angle;
- selecting one or more target bounding boxes from among the plurality of bounding boxes based on the confidence score associated with each of the plurality of bounding boxes by iteratively performing a non-maximum suppression (NMS) routine based on the confidence score associated with each of the plurality of bounding boxes until the one or more target bounding boxes do not overlap;
- and
- identifying the one or more objects based on the selected one or more target bounding boxes.
Dependencies are inferred from the claim text.