Vehicle Object Detection With Transformer-Based Height Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for object detection and height estimation in vehicle environments face challenges due to high computational and memory requirements of 3D grid maps and inaccurate elevation information from radar sensors, particularly with strong multi-path effects.
Innovation Solution
A computer-implemented method using an adapted Natural Language Processing (NLP) transformer and regression algorithm to process 3D point clouds from perception sensors, transforming spatial information into a vehicle coordinate system, clustering objects, and determining physical extensions like height, width, and depth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D grid maps are used for object detection and height estimation, then measurement precision is improved, but device complexity and memory requirements increase significantly
Solution Approach 1:
The patent segments the 3D point cloud data into multiple clusters representing different objects, then processes each cluster independently through the NLP transformer. This segmentation allows the system to maintain high measurement precision for height estimation while reducing overall computational complexity by focusing processing resources on distinct object regions rather than the entire environment map.
Solution Approach 2:
The patent creates a simplified representation of spatial information by transforming point cloud data into a format that can be processed by the NLP transformer. This copying approach allows the system to retain essential spatial information for accurate height measurement while avoiding the computational burden of full 3D grid map processing.
2Device complexity
If radar sensors are used for 3D environment perception, then device complexity is reduced, but measurement precision deteriorates due to inaccurate elevation angles and multi-path effects
Solution Approach 1:
The patent introduces an NLP transformer as an intermediary processing layer between the radar sensor and the object detection algorithm. This transformer processes the point cloud data and generates enhanced spatial representations that compensate for radar's inaccurate elevation measurements. The intermediary effectively translates imprecise radar data into accurate object height information without requiring complex sensor fusion or post-processing.
Solution Approach 2:
The patent changes the representation parameters of spatial information by transforming raw radar point cloud data into a format optimized for NLP processing. This parameter transformation allows the system to maintain the simplicity of radar sensors while achieving high measurement precision through the NLP transformer's ability to infer accurate dimensions from transformed data representations.
3Ease of operation
If classical grid mapping is used, then ease of operation is improved, but productivity decreases due to high computational and memory requirements
Solution Approach 1:
The patent replaces the traditional mechanical/grid-based mapping approach with an NLP transformer-based system. This substitution maintains ease of operation by providing simple object detection functionality while dramatically improving productivity. The NLP transformer processes point cloud data efficiently, reducing computational time and memory requirements compared to classical grid mapping methods.
Solution Approach 2:
The patent transitions from 2D/3D grid map representations to a point cloud-based approach processed by the NLP transformer. This dimensional change allows the system to maintain operational simplicity while improving computational efficiency. The NLP transformer processes spatial information in a way that reduces the computational burden compared to traditional grid-based methods.
Data Source
Figure 1
Figure 2
AI summary
The present disclosure relates to a computer implemented method for detecting objects (20) and determining one or more physical extensions of said objects (20) in an environment around a vehicle (10) based on data of one or more perception sensors (12) of the vehicle (10). Further, the present disclosure relates to a computer system (14), the computer system (14) being configured to carry out the said computer implemented method. Additionally, the present disclosure relates to a non-transitory computer readable medium comprising instructions for carrying out said computer implemented method.