Multi-Sensor Fusion Tracking for Accurate LIDAR Object Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition systems for autonomous vehicles face challenges in accurately identifying and classifying external objects using fusion tracks from multiple sensors, leading to potential errors in obstacle detection and vehicle control.
Innovation Solution
An object recognition apparatus and method that utilizes a combination of LIDAR, camera, and radar sensors to identify and classify fusion tracks, with specific instructions for identifying target fusion tracks and associating them with corresponding LIDAR points to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple sensors (camera, radar, LIDAR) are used for fusion tracking to improve object detection accuracy, then the reliability of object identification is improved, but the device complexity increases
Solution Approach 1:
The patent combines data from multiple sensors (camera, radar, LIDAR) to create fusion tracks that integrate information from different sensing modalities. This merging approach improves object identification reliability by cross-validating detections across sensors while managing complexity through coordinated processing of fused data streams.
Solution Approach 2:
The fusion track system serves multiple functions: it identifies objects, classifies them by category, determines their positions, and provides data for vehicle control decisions. This multi-functionality consolidates what would otherwise require separate processing systems, improving reliability without proportionally increasing complexity.
2Measurement precision
If fusion tracks are used to identify external objects through multiple sensors, then the accuracy of position identification is improved, but the difficulty of detecting and measuring increases
Solution Approach 1:
The patent segments the object detection process into distinct stages: individual sensor detection, fusion track generation, category classification, and position identification. This segmentation allows each stage to be optimized independently, improving measurement precision while managing processing difficulty through modular architecture.
Solution Approach 2:
The fusion track acts as an intermediary data structure that consolidates information from multiple sensors before final object identification. This intermediary layer simplifies the detection and measurement process by providing a unified representation that integrates multi-sensor data, reducing the complexity of directly processing raw sensor inputs.
3Measurement precision
If LIDAR points are associated with fusion tracks through multiple conditions (position, angle, distance), then the classification accuracy of target fusion tracks is improved, but the loss of time in processing increases
Solution Approach 1:
The patent applies position, angle, and distance conditions as preliminary filters before final classification. By pre-filtering LIDAR points that satisfy these geometric conditions, the system reduces the candidate set for classification, improving accuracy while minimizing processing time through early elimination of non-matching points.
Solution Approach 2:
The system applies multiple conditions (position, angle, distance) that may be more stringent than strictly necessary, but this excessive filtering ensures high classification accuracy. The additional computational overhead is justified by the significant improvement in classification precision, particularly for safety-critical object identification.
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
The proposed solution improves the accuracy of external object position identification and classification, reducing the frequency of erroneous vehicle control actions and enhancing the reliability of obstacle detection systems.
Implementation Method 1
A distance from a LIDAR to an object may be obtained through an interval between the time when a laser is transmitted by the LIDAR and the time when the laser reflected by the object is received
Implementation Method 2
at least one fusion track corresponding to at least one external object acquired through the camera and the radar
Data Source
AI summary
An embodiment object recognition apparatus includes a LIDAR, a camera, a radar, one or more processors, and a storage device storing a program to be executed by the one or more processors, the program including instructions for identifying a fusion track corresponding to an external object acquired through the camera and the radar, identifying a target fusion track, identifying a LIDAR track corresponding to an external object represented by the target fusion track, identifying a target LIDAR track, identifying second LIDAR points that satisfy a position condition, identifying a third LIDAR point that satisfies a distance condition identified based on a distance from a point corresponding to a host vehicle to the second LIDAR point, and associating and storing the target fusion track and the third LIDAR point based on identifying the third LIDAR point.