Multi-Camera Object Detection Fusion for V2X Blind Spot Coverage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detectors in autonomous vehicles face challenges in accurately detecting objects in actual situations, especially in blind spots and occluded areas, and often produce inconsistent results due to varying learned parameters, lacking sufficient training data, and are unable to integrate information from nearby cameras effectively.
Innovation Solution
A method and device for merging object detection information from nearby cameras using object detectors, involving a merging device and processor to combine and validate detection data through reliability ordering, matching operations, and auto-labeling to generate accurate and comprehensive object detection information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional object detectors use learned parameters for detection, then detection speed is maintained, but detection accuracy in actual situations cannot be confirmed and may be unreliable
Solution Approach 1:
The patent merges detection results from multiple object detectors located in nearby vehicles to compensate for individual detector limitations. By combining detections from multiple sources, the system achieves more reliable and accurate object detection without requiring each individual detector to be perfectly accurate, thus resolving the contradiction between reliability and device complexity.
Solution Approach 2:
The system implements feedback mechanisms where detection results are shared across the network and used to validate and improve individual detector performance. Through iterative validation and re-training using shared ground truth data, the system continuously improves detection accuracy while maintaining manageable complexity through automated processes.
2Loss of information
If multiple object detectors are deployed in nearby vehicles, then detection coverage is improved, but information integration and validation become complex
Solution Approach 1:
The system employs automated algorithms for matching detection results, generating ground truth labels, and validating detections across multiple vehicles. These self-service mechanisms handle the integration complexity automatically, allowing the system to benefit from multiple detectors without proportionally increasing manual intervention requirements.
Solution Approach 2:
The patent introduces an intermediary matching and validation system that coordinates information between multiple object detectors. This intermediary layer handles the complex task of integrating detections, resolving conflicts, and generating consistent ground truth, thereby managing integration complexity while maximizing detection coverage.
3Adaptability or versatility
If object detectors rely on fixed learned parameters, then system operation is simple, but they cannot adapt to various actual situations and blind spots
Solution Approach 1:
The system dynamically adapts detector parameters through iterative re-training using ground truth data generated from merged detection results. Rather than relying on fixed parameters, the detectors continuously evolve to adapt to various actual situations, including blind spots and occluded scenes, while the automated nature of this process maintains operational simplicity.
Solution Approach 2:
The system performs preliminary actions by pre-sharing detection results and generating ground truth labels before formal validation and re-training cycles. This preliminary preparation enables faster adaptation to new situations without requiring complex real-time adjustments, maintaining operational simplicity while improving adaptability.
4Reliability
If separate monitoring is implemented for conventional object detectors, then detection accuracy can be confirmed, but system complexity and operational burden increase
Solution Approach 1:
The system implements self-service validation where object detectors automatically validate each other's detections through mutual comparison and cross-verification. This distributed validation approach confirms detection accuracy without requiring centralized monitoring, thereby reducing operational burden while maintaining high reliability.
Solution Approach 2:
The patent establishes feedback loops where detection results are automatically shared, compared, and used to validate individual detector performance. This automated feedback mechanism provides continuous validation without manual intervention, reducing operational burden while ensuring reliable detection through multiple verification passes.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for merging object detection information detected by object detectors, each of which corresponds to each of cameras located nearby, by using V2X-based auto labeling and evaluation, wherein the object detectors detect objects in each of images generated from each of the cameras by image analysis based on deep learning is provided. The method includes steps of: if first to n-th object detection information are respectively acquired from a first to an n-th object detectors in a descending order of degrees of detection reliabilities, a merging device generating (k-1)-th object merging information by merging (k-2)-th objects and k-th objects through matching operations, and re-projecting the (k-1)-th object merging information onto an image, by increasing k from 3 to n. The method can be used for a collaborative driving or an HD map update through V2X-enabled applications, sensor fusion via multiple vehicles, and the like.