Vehicle Bounding Box Assignment via Velocity-Based Subset Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automotive perception systems struggle to provide precise bounding boxes for large vehicles due to biased size distribution and great variance in object sizes, leading to multiple small bounding boxes being assigned to one large vehicle.
Innovation Solution
A method and device that acquire sensor data from radar and Lidar sensors to determine preliminary bounding boxes, estimate their velocities, and select subsets based on velocity to merge preliminary bounding boxes into a final bounding box, effectively addressing the issue of size variance and bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data balancing is applied to down-sample small vehicles and over-sample large vehicles, then the bias toward small vehicles is reduced, but multiple small bounding boxes are still assigned to large vehicles in many cases
Solution Approach 1:
The patent changes the parameter used for selecting preliminary bounding boxes from spatial proximity alone to velocity-based selection. By using velocity information, the system can distinguish between multiple small vehicles and parts of a single large vehicle, thereby improving bounding box assignment accuracy while maintaining precision for large vehicles
Solution Approach 2:
The patent introduces velocity as an intermediary parameter to mediate the selection process. Velocity information acts as a mediator that helps determine whether multiple preliminary bounding boxes belong to the same object, resolving the contradiction between reliability and precision in bounding box assignment
2Productivity
If machine learning models are trained with biased size distribution of vehicles, then the system processes data efficiently, but small bounding boxes are assigned to any type of vehicles including large ones
Solution Approach 1:
The patent applies preliminary action by estimating velocities of preliminary bounding boxes before the final selection process. This preliminary velocity estimation allows the system to correct potential errors in bounding box assignment caused by biased training data, thereby improving measurement precision without sacrificing processing efficiency
Solution Approach 2:
The patent implements feedback by using velocity information to validate and correct the output of machine learning models. The velocity-based selection process provides feedback that counteracts the bias introduced during training, ensuring accurate bounding box assignment for vehicles of all sizes while maintaining efficient data processing
Data Source
AI summary
A method is provided for assigning a bounding box to an object in an environment of a vehicle. Data related to objects located in the environment of the vehicle are acquired via a sensor. Based on the data, a respective spatial location and a respective size of a plurality of preliminary bounding boxes are determined such that each preliminary bounding box covers one of the objects at least partly. A respective velocity of each preliminary bounding box is estimated based on the data. A subset of the plurality of preliminary bounding boxes being related to a respective one of the objects is selected, where the subset is selected based on the respective velocity of each of the preliminary bounding boxes. A final bounding box is assigned to the respective one of the objects by merging the preliminary bounding boxes of the corresponding subset.