Bounding Curve Feature Extraction for Autonomous Vehicle Object Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Autonomous vehicles face challenges in accurately classifying objects in their environment, which can impact safe navigation and control.

Innovation Solution

A system and method that utilize a processor to determine bounding curve features from sensor data, applying them to a machine learning model for object classification, including convexities and concavities, to assist in autonomous vehicle control.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional object classification methods are used in autonomous vehicles, then the system complexity is lower, but the object classification accuracy is insufficient

Engineering Contradiction:

Improveobject classification accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the object classification process into distinct stages: sensor data acquisition, bounding curve generation, convexity/concavity feature extraction, and machine learning-based classification. This segmentation allows each component to be optimized independently, improving overall classification accuracy while managing system complexity through modular architecture.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent transitions from traditional 2D image-based classification to 3D point cloud processing with bounding curves that capture spatial geometry. By adding the dimension of 3D spatial relationships and geometric features (convexities and concavities), the system achieves superior classification accuracy for autonomous vehicle applications.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Manufacturing precision

If simple bounding box methods are used, then the processing speed is faster, but the geometric feature extraction precision is insufficient

Engineering Contradiction:

Improvegeometric feature extraction precisionVSAvoidprocessing speed

Core Design Contradiction:

Manufacturing precisionVSSpeed

Solution Approach 1:

The patent employs bounding curves that follow the actual contours of objects, capturing curved surfaces and complex geometries. By using curvature-based representations instead of straight-line bounding boxes, the system extracts precise geometric features including convexities and concavities, enabling better differentiation of object shapes while maintaining processing efficiency through optimized curve algorithms.

Inventive Principle:

Principle #14Spheroidality (Curvature)

Data Source

PatentUS10430673B2Systems and methods for object classification in autonomous vehicles

Publication Date: 2019.10.01 GM GLOBAL TECHNOLOGY OPERATIONS LLC

AI summary

Systems and method are provided for controlling a vehicle. In one embodiment, an object classification method includes receiving sensor data associated with an object observed by a sensor system of an autonomous vehicle and determining, with a processor, a bounding curve associated with the sensor data. A plurality of bounding curve features are determined based on a set of convexities and concavities associated with the bounding curve. The object is classified by applying the plurality of bounding curve features to a machine learning model and receiving a classification output.