3D Bounding Shape Representation for Multi-Axis Object Orientation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches to object detection using LiDAR-camera fused bird's-eye view (BEV) datasets are inadequate for environments where sensors are mounted on fixed structures, leading to inaccurate representation of object orientations and inefficient use of computing resources due to unnecessary data processing.
Innovation Solution
Generate extended bounding shape representations using multiple orientation parameters (9 degrees of freedom) for objects, including rotations about the x, y, and z-axes, through a machine learning model trained with synthetically generated ground truth data, to accurately define bounding shapes in environments with static sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single rotation parameter is used to represent bounding box orientation, then the representation is simple and computationally efficient, but the orientation accuracy is insufficient for objects with complex spatial arrangements
Solution Approach 1:
The patent extends the bounding box representation from a single rotation parameter (one dimension) to multiple rotation parameters representing rotations about three different axes (three dimensions). This dimensional expansion enables the system to capture complex object orientations and spatial arrangements that cannot be represented by a single angle, thereby improving measurement precision without excessive complexity increase.
2Productivity
If conventional bird's-eye view fusion is used for object detection, then the processing is efficient for autonomous driving, but unnecessary data processing occurs in environments with fixed structure-mounted sensors
Solution Approach 1:
The patent changes the parameter representation from a single rotation angle to multiple rotation parameters corresponding to rotations about three axes. This parameter transformation enables the system to adapt to different sensor mounting configurations (including fixed structures) without requiring complete reprocessing of the entire dataset, thereby reducing unnecessary computational energy while maintaining detection accuracy.
3Reliability
If multiple rotation parameters are used to represent bounding box orientation, then the orientation representation is comprehensive and accurate, but the computational overhead increases
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing rotation matrices corresponding to multiple rotation parameters during system initialization or offline processing. During real-time operation, the system can directly retrieve and apply these pre-computed matrices rather than performing complex real-time calculations, thereby reducing computational resource consumption while maintaining high reliability in object identification.
Data Source
AI summary
In various examples, embodiments are directed to generating extended bounding shape representations corresponding with objects in an environment in an efficient and effective manner. In particular, a bounding shape associated with an object may be represented using various parameters, including position parameters, dimension parameters, and orientation parameters that describe the spatial properties of an object. Advantageously, the orientation parameters include representations or indications of rotation about an x-axis, a y-axis, and a z-axis. Orientation parameters associated with multiple orientations, such as angles of rotations about the x-axis, the y-axis, and the z-axis, facilitate a more comprehensive analysis of an environment, particularly in instances in which sensors, such as a camera and LiDAR, are mounted on a wall or ceiling or in other instances in which rotation angles may exist in association with multiple axes.