Vehicle Appendage Labeling From LiDAR for Accurate Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately identifying and labeling vehicle appendages, such as side mirrors or ramps, using lidar data, which can lead to incorrect spatial location and unsafe operation due to the large volume of lidar data and the complexity of object recognition.
Innovation Solution
A method is developed to train a perception model using machine learning algorithms to identify and label vehicle appendages by processing lidar points, generating bounding boxes around both the vehicle and its appendages, and associating them correctly, employing a taxonomy for semantic labeling and loss functions to ensure accurate classification and relationship recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional object recognition methods are used to process lidar data, then the processing speed is maintained, but the accuracy of identifying vehicle appendages deteriorates due to the large volume of data and complexity of recognition
Solution Approach 1:
The patent segments the object recognition task into two distinct stages: first identifying the parent vehicle, then identifying appendages associated with that vehicle. This segmentation allows each stage to focus on specific features, improving accuracy while managing complexity. The loss function is also segmented to separately optimize vehicle detection and appendage detection.
Solution Approach 2:
The patent introduces an intermediary relationship model that connects vehicles and appendages through semantic labels and spatial relationships. This intermediary structure allows the system to reason about appendages in the context of their parent vehicles, improving identification accuracy without requiring direct complex processing of all lidar points simultaneously.
2Measurement precision
If the perception model processes all lidar points to identify both vehicles and appendages, then the identification accuracy improves, but the computational time and resources increase
Solution Approach 1:
The system performs preliminary identification of parent vehicles before proceeding to identify appendages. By first establishing the vehicle detection results and then using those results to guide appendage search, the system avoids redundant processing and reduces overall computational time while maintaining comprehensive identification accuracy.
Solution Approach 2:
The patent applies partial action by focusing computational resources on regions and features relevant to appendage detection only after vehicle identification is complete. Rather than processing all lidar points with equal intensity, the system concentrates computational effort where it is most needed, reducing overall processing time while maintaining accuracy.
Data Source
AI summary
The present technology is directed to identifying and labeling a vehicle appendage. More specifically, the present technology is generally related to receiving one or more lidar points associated with a vehicle having an appendage and outputting a label to classify the vehicle and the appendage. In some examples, a first portion of the one or more lidar points associated with the vehicle having the appendage represents the vehicle and a second portion of the one or more lidar points associated with the vehicle having the appendage represents the appendage The present disclosure can further train a perception model to output the label to classify the vehicle and the appendage.