Object Recognition Using LIDAR Lane Geometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition systems for autonomous vehicles face challenges in accurately determining the position of objects with respect to a host lane, especially on curved paths and at varying distances, which can lead to reduced accuracy and increased risk of accidents.
Innovation Solution
The proposed object recognition apparatus and method utilize a LIDAR and processors to determine the position of objects by calculating contour points, identifying left and right lines of the host lane, and determining straight lines based on reference points within an object box, thereby improving positional accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional object recognition methods are used to determine object position relative to host lane, then the system structure remains simple, but the positioning accuracy deteriorates on curved paths and at varying distances
Solution Approach 1:
The patent segments the object box into multiple reference points (first reference point and second reference point) and processes different straight lines (first straight line, second straight line, third straight line) separately. Each straight line is determined based on specific geometric relationships with the host lane boundaries, allowing independent calculation of positional relationships that can be combined for improved accuracy.
Solution Approach 2:
The patent transitions from two-dimensional bounding box representation to three-dimensional spatial reasoning by introducing multiple reference points within the object box and calculating their relationships with the host lane in 3D space. This enables more accurate positioning by considering both longitudinal and lateral dimensions simultaneously.
2Measurement precision
If simple reference point methods are used for position determination, then the calculation process remains fast, but the accuracy deteriorates on curved paths
Solution Approach 1:
The patent performs preliminary determination of the host lane boundaries (left line and right line) and establishes the coordinate system before processing object positions. The straight lines representing host lane boundaries are pre-calculated based on the curvature, allowing subsequent object positioning to use these pre-established references rather than recalculating from scratch.
Solution Approach 2:
The patent applies different calculation methods for different parts of the object box. The first reference point (e.g., center point) is used for one straight line determination, while the second reference point (e.g., corner point) is used for another, allowing each point to contribute optimally to the overall positioning accuracy based on its local geometric properties.
3Reliability
If single reference point method is used, then the processing remains simple, but the reliability of position determination deteriorates at varying distances
Solution Approach 1:
The patent merges information from multiple reference points (first reference point and second reference point) and multiple straight lines (first, second, and third straight lines) to determine the final object position. By combining the results from different geometric relationships and reference points, the system achieves more reliable positioning that is less susceptible to errors at varying distances.
Solution Approach 2:
The patent uses the determined straight lines and reference points to calculate the positional relationship between the object and host lane, then uses this feedback information to refine the position determination. The system continuously validates the position by checking consistency across multiple reference points and straight lines, improving reliability through iterative verification.
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
This approach enhances the accuracy of object positioning relative to the host lane, reduces the risk of accidents, and improves vehicle operation comfort by effectively handling curved paths and varying distances.
Implementation Method 1
A distance from a LIDAR to an object may be obtained through an interval between the time when a laser light is transmitted by the LIDAR and the time when the laser light reflected by the object is received
Data Source
AI summary
An embodiment object recognition apparatus includes a LIDAR, one or more processors, and a non-transitory storage storing a program to be executed by the one or more processors and including instructions to determine left and right lines located on left and right sides of a host lane, determine a first straight line based on a first reference point included in an object box and the left line, a second straight line based on the first reference point and the right line, a third straight line based on the first and second straight lines and representing a middle line of the host lane, or any combination thereof, and determine a state corresponding to a position of the object with respect to the host lane based on a relative positional relationship between one or more of the first, second, and third straight lines and a second reference point representing a position of the object box.