3D Bounding Box Association Using 2D Image Characteristics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in accurately associating three-dimensional bounding boxes with tracked objects in autonomous vehicles, leading to incorrect associations due to factors like relative position changes and occlusions.
Innovation Solution
A system that generates three-dimensional bounding boxes using two-dimensional image characteristics, including anchor points, pixel information, and neural network analysis, to improve the association process by incorporating two-dimensional image data for tracking and correcting errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If three-dimensional bounding boxes are generated from two-dimensional images using neural network models, then object detection capability is improved, but association accuracy with tracked objects deteriorates due to incorrect associations
Solution Approach 1:
The patent introduces two-dimensional image characteristics (such as color, texture, shape features extracted from the 2D image) as an intermediary to bridge the gap between 3D bounding box detection and tracked object association. These 2D characteristics serve as additional identifiers that help correctly match detected objects with their corresponding tracked objects, resolving the association accuracy problem while maintaining the benefits of neural network-based detection.
2Ease of operation
If only three-dimensional bounding box information is used for object tracking, then tracking simplicity is maintained, but tracking precision deteriorates due to incorrect associations in dense traffic scenarios
Solution Approach 1:
The patent merges two-dimensional image characteristics with three-dimensional bounding box information to create a composite tracking system. By combining 2D visual features (color, texture, shape) with 3D spatial information, the system achieves higher tracking precision in dense traffic scenarios while maintaining operational simplicity through automated feature extraction and association algorithms.
Data Source
AI summary
Systems and methods for associating three-dimensional bounding boxes with tracked objects are disclosed based on information gathered from objects in a two-dimensional image, for applications such as autonomous navigation. The systems and methods track an object in three-dimensional space and receive a two-dimensional image from a vehicle sensor. The system generates a three-dimensional bounding box for an object in the two-dimensional image, determines a two-dimensional image characteristic of the object, and associates the three-dimensional bounding box with the tracked object based on the three-dimensional bounding box of the object and the two-dimensional image characteristic of the object.