Autonomous Vehicle Trajectory Prediction via Object Interaction Database
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving technologies face challenges in accurately predicting trajectories of objects around a main vehicle, leading to potential traffic accidents due to errors in decision-making.
Innovation Solution
A method and electronic apparatus that utilize a processor to receive video frames, perform object recognition, retrieve preset interactive relationship information from a database, and determine a trajectory for the main vehicle based on these interactions, thereby generating a predicted trajectory for navigating the vehicle while considering the interactions between objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional trajectory prediction methods are used without considering object interactions, then the computational complexity is low, but the trajectory prediction accuracy is poor leading to safety issues
Solution Approach 1:
The system performs preliminary object recognition and interaction relationship establishment before trajectory prediction. By pre-processing the scene to identify objects and their interactions, the system prepares necessary information in advance, improving prediction accuracy without proportionally increasing real-time computational complexity
Solution Approach 2:
The patent introduces an interaction relationship database as an intermediary between object detection and trajectory prediction. This database stores pre-defined interaction patterns between different object types, serving as a knowledge base that guides trajectory prediction without requiring complex real-time calculations of all possible interactions
2Reliability
If more objects and their interactions are analyzed in real-time, then the decision-making accuracy improves, but the processing time increases
Solution Approach 1:
The system applies different processing depths to different objects based on their relevance to the main vehicle. Objects with higher interaction potential (e.g., pedestrians near the vehicle, vehicles in adjacent lanes) receive more detailed analysis, while distant or less relevant objects receive simplified processing, optimizing the balance between accuracy and processing time
Solution Approach 2:
Interaction relationship databases and object recognition models are pre-trained and stored, allowing the system to quickly query pre-computed interaction patterns during real-time operation rather than calculating all interactions from scratch, significantly reducing processing time while maintaining high accuracy
Data Source
AI summary
A method and an electronic apparatus for predicting a path based on an object interaction relationship are provided. The method includes the following. A video including a plurality of image frames is received. Object recognition is performed on a certain image frame among the plurality of image frames to recognize at least one object in the certain image frame. Preset interactive relationship information associated with the at least one object is obtained from an interactive relationship database based on the at least one object. A first trajectory for navigating the first vehicle is determined based on the preset interactive relationship information.