Multiscale Graph Convolutional Network for Road Intersection Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for automatic intersection detection, such as heuristic and machine learning approaches, face challenges in accurately identifying road intersections from large-scale GPS trajectories, particularly in areas with incomplete or inaccurate map data, leading to suboptimal routing decisions and inaccurate arrival time estimations.
Innovation Solution
A method utilizing a multiscale graph convolutional network that determines node vectors from geographical data, including GPS data, and inputs them into a trained classifier to predict the probability of a location being a road intersection, effectively learning from high-quality map data to detect intersections in areas with poorer data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If heuristic methods or traditional machine learning methods are used for intersection detection, then the implementation is simpler, but the detection accuracy deteriorates in areas with incomplete or inaccurate map data
Solution Approach 1:
The patent transforms the intersection detection problem from a geometric/mapping-based task into a classification task by changing the parameters used for detection. Instead of relying on map data completeness, the system uses GPS trajectory patterns and temporal features (e.g., stopping behavior, direction changes) as input parameters for a trained classifier, enabling accurate detection even in areas with poor map data
Solution Approach 2:
The patent replaces traditional geometric intersection detection methods (which rely on map data and spatial relationships) with a machine learning-based classification system. The classifier learns from labeled GPS trajectory data and temporal features, substituting the mechanical/geometric detection approach with an intelligent system that can generalize to areas with incomplete map information
2Measurement precision
If more complex machine learning models are used to improve detection accuracy, then the intersection detection accuracy improves, but the computational complexity and training requirements increase
Solution Approach 1:
The patent segments the intersection detection process into distinct components: feature extraction (temporal features from GPS data), model training (using labeled trajectories), and classification (applying the trained model to new data). This segmentation allows for targeted optimization of each component while managing overall system complexity
Solution Approach 2:
The patent performs preliminary actions by pre-training the classifier model using large-scale labeled GPS trajectory data before deployment. This preliminary training phase captures the complex patterns of intersection behavior, so that during actual detection, the system can make accurate predictions without performing complex real-time computations
3Use of energy by moving object
If traditional methods are used, then the computational resources required are lower, but the ability to generalize to areas with poor map data deteriorates
Solution Approach 1:
The patent enables the system to serve itself by learning from the data it processes. The classifier is trained on GPS trajectory data and automatically adapts to different geographic areas and map data qualities without requiring manual configuration or reprogramming for each region
Solution Approach 2:
The system incorporates feedback mechanisms where the classifier learns from labeled trajectory data and improves its performance over time. The model uses feedback from training examples to adjust its parameters and generalize better to areas with poor map data, creating a self-improving system
Data Source
AI summary
A method, data processing apparatus, and computer code for identifying road intersections includes providing location data obtained from at least one vehicle's trajectory, wherein the location data may include geographical data within a geographical perimeter. The method includes determining node vectors by applying a geographical descriptor model on a target location included in the geographical perimeter. The geographical descriptor model includes a plurality of multiscale node descriptors including a target multiscale descriptor and neighboring multiscale descriptors. Each of the plurality of multiscale node descriptors includes at least two shape descriptors of different geographical resolution. Each of the neighboring locations is at a respective geographical distance from the target location. The node vectors may be respectively determined for each of the plurality of multiscale node descriptors. The method includes inputting the node vectors into a trained multiscale classifier including a graph convolutional network to provide a probability of the target location being a road intersection.