3D Object and Lane Detection Association Using Graph Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wireless communication systems, particularly 5G NR, face challenges in improving the accuracy and reliability of object detection tasks, as polylines and 3D object detection are typically treated independently, leading to inaccuracies and false positives.
Innovation Solution
The integration of polyline and 3D object detection models using a graph neural network (GNN) to associate and update outputs, capturing long-term dependencies, thereby enhancing the accuracy and robustness of both tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If polyline detection and 3D object detection are performed independently, then the detection process is simple and fast, but the accuracy and reliability of object detection deteriorates due to false positives and inaccuracies
Solution Approach 1:
The patent merges polyline detection and 3D object detection into a unified detection framework using a single neural network model. The model simultaneously outputs both polyline predictions and 3D object detections, allowing the two detection tasks to inform and correct each other. This integration resolves the contradiction by improving reliability through cross-validation while managing complexity through a unified architecture rather than separate systems.
Solution Approach 2:
The detection model is designed with multi-functionality, serving both as a polyline detector and a 3D object detector. The same model structure and training process handle both detection tasks, enabling the system to perform multiple functions simultaneously. This universality improves reliability by allowing one system to validate both lane boundaries and objects, while avoiding the complexity of maintaining separate specialized systems.
2Measurement precision
If polyline detection and 3D object detection are performed independently, then the computational resources required are reduced, but the localization accuracy deteriorates
Solution Approach 1:
By combining both detection tasks into a single neural network, the system achieves better localization accuracy through shared computational resources. The unified model processes input data once and produces both polyline and object detection results, improving measurement precision while avoiding the redundant computational energy consumption that would result from running separate detection systems.
3Reliability
If separate detection models are used for polylines and 3D objects, then the system is easier to implement and maintain, but the overall detection accuracy decreases due to lack of association between polylines and objects
Solution Approach 1:
The patent implements a unified detection model that simultaneously performs polyline detection and 3D object detection, establishing associations between the two detection results. This merging approach improves reliability by enabling cross-validation and spatial reasoning between polylines and objects, while the model is designed to maintain ease of implementation through a single trained network rather than multiple separate systems requiring coordination.
Data Source
AI summary
Aspects presented herein may improve the accuracy and reliability of object detections performed by multiple object detection models. In one aspect, a UE detects (1) a set of polylines from at least one of a set of bird's eye view (BEV) features or a set of perspective view (PV) features associated with a set of images and (2) a set of three-dimensional (3D) objects in the set of BEV features. The UE associates the set of polylines with the set of 3D objects. The UE updates the set of polylines based on a set of nearby 3D objects or updates the set of 3D objects based on a set of nearby polylines. The UE outputs an indication of the updated set of polylines or the updated set of 3D objects.