Autonomous Vehicle Object Classification Using LiDAR Box Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle control systems face inaccuracies in determining the type of external objects using LiDAR-based virtual boxes and neural network models, leading to incorrect classifications.
Innovation Solution
A vehicle control apparatus and method that utilize a processor to obtain and process point clouds from LiDAR, generate first and second virtual boxes, and update classification information based on association, reliability, and overlap criteria to improve object type determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the type of external object is determined using a virtual box based on LiDAR point cloud, then the object classification can be performed, but the classification accuracy may be incorrect
Solution Approach 1:
The patent combines two different classification approaches: traditional virtual box-based classification and neural network-based classification. The processor generates both a first virtual box using traditional methods and a second virtual box using a neural network model, then merges these results through association determination to produce a final classification decision, thereby improving overall classification accuracy
Solution Approach 2:
The system implements feedback by comparing the first virtual box results with the second virtual box results from the neural network. The processor determines whether to update the first classification information based on the association between the two virtual boxes, creating a feedback loop that refines classification accuracy by validating results across multiple methods
2Reliability
If multiple classification methods are used to improve accuracy, then the classification reliability improves, but the system complexity increases
Solution Approach 1:
The system performs preliminary classification using the traditional virtual box method to generate first classification information before applying the more complex neural network method. This preliminary action allows the system to quickly filter and prepare data, reducing the overall computational burden while maintaining reliability through the subsequent neural network validation
Solution Approach 2:
The patent applies different classification methods to different situations by determining association between virtual boxes. The system selectively updates classification information based on the neural network results only when association is determined, applying complexity locally rather than universally, thereby balancing reliability improvement with system complexity management
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances the accuracy of external object classification, thereby improving the performance of autonomous driving systems by ensuring precise decision-making in driving assistance and autonomous modes.
Implementation Method 1
the vehicle may identify the external object by using a sensor (e.g., a light detection and ranging (LiDAR))
Data Source
AI summary
An apparatus for controlling autonomous driving of a vehicle may comprise a sensor, a memory configured to store a neural network model, and a processor. The processor is configured to obtain a first virtual box based on a cluster of points representing an external object detected by the sensor, and a second virtual box by inputting the cluster of points into the neural network model. The processor initializes second classification information among first and second classification information included in the first virtual box. The processor determines whether to update the second classification information using the second virtual box based on the association between the first and second virtual boxes. The processor then outputs the first virtual box by assigning either the first or second classification information, generates a signal indicating the assigned classification information, and controls the autonomous driving of the vehicle based on the signal.