Moving Object Prediction Using Presence Probability Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing moving object prediction technologies face challenges in accurately predicting the future position of moving objects due to unclear movement limitations, reliance on driver operations, incomplete coverage of traffic scenarios, and lack of physical significance in risk evaluation.
Innovation Solution
A system that detects a moving object's position and behavior state, generates a presence probability map, and uses particle distributions to predict future positions based on movement states and presence probabilities, covering various conditions and scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If movement limitations are not clearly defined, then the moving object can move anywhere, but prediction precision of future position deteriorates
Solution Approach 1:
The patent applies local quality by creating a presence probability map where different spatial regions have different probability values indicating likelihood of object presence. This allows the system to constrain prediction to high-probability regions while maintaining flexibility elsewhere, resolving the contradiction between movement freedom and prediction precision.
Solution Approach 2:
The patent uses dynamic particle distributions that evolve over time based on movement state distributions. The particles are dynamically adjusted according to presence probabilities in different regions, allowing the system to adaptively constrain predictions to plausible areas while maintaining flexibility for unexpected movements.
2Reliability
If risk evaluation is based on designer scoring, then the system can evaluate risk, but physical significance is lost
Solution Approach 1:
The patent replaces the mechanical/scoring-based risk evaluation system with a probabilistic field-based system. Instead of designer-defined scores, the system uses presence probability maps and movement state distributions that have clear physical significance in terms of spatial probability density, preserving physical meaning while maintaining evaluation capability.
3Reliability
If database of accident states is required, then collision prevention can be achieved, but all actual situations cannot be covered
Solution Approach 1:
The patent creates a universal prediction framework using presence probability maps and particle distributions that can handle any traffic situation without requiring specific database entries. The system universally applies probability-based reasoning to novel situations, achieving both reliability and adaptability.
Solution Approach 2:
The patent performs preliminary action by pre-computing presence probability maps for various regions before actual prediction is needed. This allows the system to quickly evaluate novel situations by referencing pre-computed probability fields rather than searching databases, improving both coverage and efficiency.
Data Source
Figure 1
Figure 2A~2C
Figure 3
AI summary
A position, behavior state and movement state of a moving object are detected, together with plural categories of track segment region and stationary object regions, using an environment detection section (40). A presence probability is applied to the detected track segment regions and stationary object regions and a presence probability map is generated, using a map generation section (42). A moving object position distribution and movement state distribution are generated by a moving object generation section (44) based on the detected moving object position, behavior state and movement state, and recorded on the presence probability map. The moving object position distribution is moved by a position update section (46) based on the moving object movement state distribution. The moved position distribution is changed by a distribution change section (48) based on the presence probabilities of the presence probability map, and a future position distribution of the moving object is predicted on the presence probability map. Consequently, the future position of the moving object can be predicted with good precision under various conditions.