Intersection Region Detection Without HD Maps for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for autonomous driving struggle to accurately and efficiently detect and classify intersection contention areas, especially in complex urban environments, due to reliance on detailed HD maps and the inability to handle occlusions and transient conditions.
Innovation Solution
The use of a deep neural network (DNN) to process sensor data from vehicles, enabling real-time or near real-time detection and classification of intersection contention areas without prior knowledge or experience of the intersection, and without relying on HD maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional systems use multiple separate deep neural networks to detect and combine various features (traffic lights, stop signs, vehicle positions, lanes, free-space boundaries) for intersection detection, then the detection accuracy may be improved, but the device complexity and computational requirements increase significantly
Solution Approach 1:
The patent combines multiple separate deep neural networks into a single unified deep neural network that performs all intersection-related detections simultaneously. This single network detects intersection areas, traffic control devices, vehicle positions, orientations, lanes, and free-space boundaries in one integrated system, reducing device complexity while maintaining comprehensive detection capabilities.
Solution Approach 2:
The unified deep neural network is designed to perform multiple functions within a single system. It simultaneously classifies intersection areas into multiple categories (intersection interior, intersection exterior, no-lane area, unclear area, pedestrian crossing area) and detects various features (vehicles, pedestrians, traffic lights, stop signs, lanes, boundaries) using a single multi-functional model rather than separate specialized networks.
2Measurement precision
If detailed annotations are required to train deep neural networks for complex intersection detection, then the detection accuracy improves, but the scalability of the system decreases
Solution Approach 1:
The system uses a comprehensive set of intersection area classifications that covers all possible scenarios (intersection interior, exterior, no-lane areas, unclear areas, pedestrian crossings). By providing slightly more detailed classification categories than strictly necessary, the system achieves robust accuracy across diverse intersection types while maintaining scalability through the unified network architecture that can adapt to different classification requirements.
3Reliability
If map-based solutions are used to analyze intersections by comparing detected features to pre-stored high-definition maps, then the reliability of intersection understanding improves, but the system fails when maps are outdated or unavailable
Solution Approach 1:
The unified deep neural network performs self-service by directly detecting and classifying all intersection features and areas from sensor data without requiring external map data. The system independently identifies intersection interiors, exteriors, no-lane areas, unclear areas, pedestrian crossings, vehicles, pedestrians, traffic control devices, and lanes through its own processing capabilities, making it self-sufficient and adaptable to any environment without relying on pre-stored maps that may be outdated or unavailable.
Data Source
AI summary
In various examples, live perception from sensors of a vehicle may be leveraged to detect and classify intersection contention areas in an environment of a vehicle in real-time or near real-time. For example, a deep neural network (DNN) may be trained to compute outputs—such as signed distance functions—that may correspond to locations of boundaries delineating intersection contention areas. The signed distance functions may be decoded and/or post-processed to determine instance segmentation masks representing locations and classifications of intersection areas or regions. The locations of the intersections areas or regions may be generated in image-space and converted to world-space coordinates to aid an autonomous or semi-autonomous vehicle in navigating intersections according to rules of the road, traffic priority considerations, and/or the like.