Trajectory Prediction Using 2D Image-Like Data Structures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current trajectory prediction methods for autonomous vehicles are computationally expensive and inaccurate, failing to effectively incorporate static road context and dynamic interactions, which are crucial for reliable navigation and collision avoidance.
Innovation Solution
A computer-implemented method using a 2D image-like data structure that combines lane information and dynamic vehicle data, processed through a convolutional neural network (CNN) to learn the correlation between static road features and vehicle trajectories, enabling efficient prediction of vehicle attributes such as location, speed, and collision risk.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If commonly used trajectory prediction methods are used, then trajectory prediction can be performed, but the computational cost is high and accuracy is low
Solution Approach 1:
The patent transforms trajectory prediction data into an image-like 2D data structure where spatial relationships and temporal sequences are represented visually. This allows applying CNN algorithms originally designed for image processing to trajectory prediction, achieving both high accuracy and computational efficiency through the dimensional transformation of data representation
Solution Approach 2:
The patent replaces traditional mechanical trajectory prediction algorithms with a neural network-based system. By substituting conventional computational methods with machine learning models, the system achieves superior accuracy while maintaining real-time processing capabilities through parallel computation inherent in neural networks
2Reliability
If static road context and dynamic interactions are incorporated, then prediction reliability is improved, but system complexity increases
Solution Approach 1:
The patent merges static road context data and dynamic vehicle interaction data into a unified image-like data structure. By combining multiple data sources into a single integrated representation, the system achieves comprehensive prediction reliability without proportionally increasing system complexity, as the unified structure allows simultaneous processing of all factors
Solution Approach 2:
The patent creates a multi-functional data structure that simultaneously represents spatial position, temporal sequence, road context, and vehicle dynamics. This universal representation format allows a single processing pipeline to handle diverse input factors, reducing overall system complexity while improving prediction reliability through comprehensive factor integration
Data Source
Figure 1A~1B
Figure 1C
Figure 2
AI summary
A computer implemented method for determining an attribute of an object at a pre-determined point comprising the following steps carried out by computer hardware components: determining data representing a plurality of properties of a plurality of items of a vicinity of the object at a plurality of points along a trajectory of the object; arranging the data in a multi-channel image-like data structure, the multi-channel image-like data structure having a plurality of columns, a plurality of rows, and a plurality of channels, wherein the data is arranged in the image-like data structure; and determining the attribute of the object at the pre-determined point using a pre-determined rule based on the image-like data structure.