Moving Track Prediction Using Road-Constrained Destination States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing moving track prediction methods in intelligent driving systems have low precision due to their reliance on kinematic models that do not account for road structures and actual motion changes, leading to significant deviations in predicting future target tracks.
Innovation Solution
A method and apparatus that incorporate preset path information, such as road structures and motion models, to generate destination states and predict moving tracks, using Frenet coordinate systems and probability calculations to improve prediction accuracy, allowing for longer-term predictions (e.g., 10 seconds) and better alignment with actual scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If track prediction is performed using only kinematic models (constant velocity, constant acceleration, or constant angular velocity) based on current location and velocity, then the prediction process is simple and fast, but the prediction precision is low due to large deviations when targets adjust motion based on road structure
Solution Approach 1:
The patent introduces road structure information as an intermediary element that mediates between the target's motion state and the predicted track. The road structure (curves, intersections, barriers) acts as a constraint that guides the target's motion, and by incorporating this intermediary information, the prediction model can account for how targets naturally adjust their paths according to road geometry, significantly improving prediction precision without requiring overly complex models
Solution Approach 2:
The patent transforms the static kinematic model into a dynamic prediction framework by introducing multiple possible future tracks with different probabilities. Instead of assuming constant motion parameters, the system dynamically adjusts predictions based on road structure constraints and target behavior patterns, allowing the model to adapt to changing motion conditions while maintaining computational efficiency
2Duration of action of moving object
If prediction time horizon is extended to long-term (e.g., 10 seconds), then more future track information is obtained for path planning, but prediction accuracy deteriorates due to accumulated deviations from simple kinematic models
Solution Approach 1:
The patent applies preliminary action by pre-establishing road structure information and constraints before performing track prediction. The system pre-processes road map data to identify curves, intersections, and barriers that will constrain target motion, and uses these pre-computed constraints to guide long-term predictions. This allows the model to maintain accuracy over extended time horizons (up to 10 seconds) by continuously referencing the pre-established road structure rather than relying solely on extrapolating current motion states
Solution Approach 2:
The patent implements feedback mechanisms by continuously comparing predicted tracks with road structure constraints and adjusting predictions accordingly. The system uses probability distributions over multiple possible tracks, where road structure information provides feedback thatprunes unrealistic trajectories and enhances likely ones, maintaining prediction accuracy even over long time horizons where simple kinematic extrapolation would accumulate significant deviations
Data Source
AI summary
A moving track prediction method includes obtaining an initial state, of a moving target that includes an initial location and an initial motion state, generating one or more destination states of the moving target based on the initial state of the moving target and preset path information, and predicting a moving track of the moving target based on the initial state of the moving target and the one or more destination states to obtain one or more predicted moving tracks.