Pedestrian Occupancy Maps for Collision Avoidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pedestrian warning systems face challenges in accurately predicting the future location of pedestrians due to unpredictable motion, leading to delays in collision warnings and increased difficulty for drivers to take evasive action.
Innovation Solution
The use of probability distribution mapping and pedestrian occupancy maps (POMs) to compactly represent the predicted future location of pedestrians, allowing for rapid and reliable transmission of location distributions, which reduces data transmission by approximately 84% and eliminates the problem of 'track gapping' by providing multiple time-phased predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional kinematic equations are used to predict object locations, then the prediction process is simple and fast, but the prediction accuracy is poor for non-kinematic objects like pedestrians
Solution Approach 1:
The patent replaces traditional kinematic equations (mechanical prediction system) with a neural network-based behavioral mapping system. The neural network learns complex pedestrian behavior patterns from training data and predicts future locations based on current state and historical behavior, achieving high accuracy for non-kinematic objects without requiring complex physics equations.
Solution Approach 2:
The system transforms the prediction approach by changing from fixed kinematic parameters (velocity, acceleration) to behavioral parameters learned from data (behavioral mode, trajectory patterns). The neural network processes input features including current position, heading, speed, and contextual information to output predicted locations that reflect actual pedestrian behavior rather than idealized motion.
2Measurement precision
If detailed object location information is transmitted to vehicles, then collision detection accuracy is improved, but data transmission time increases reducing driver response time
Solution Approach 1:
The system extracts only the essential predictive information needed for collision detection - specifically the predicted future location distribution of pedestrians - and transmits this condensed data to vehicles. Rather than transmitting all raw sensor data or detailed trajectory information, the system extracts and transmits only the critical prediction results, reducing transmission time while maintaining detection accuracy.
Solution Approach 2:
The neural network performs preliminary prediction of pedestrian future locations before transmission to the vehicle. This advance computation allows the system to prepare collision detection data in ahead of time, so that when data is transmitted to the vehicle, the processing is already complete or near-complete, minimizing the time drivers need to respond.
3Reliability
If multiple time-phased predictions are provided, then track gapping is eliminated and prediction reliability is improved, but computational load increases
Solution Approach 1:
The patent segments the prediction output into multiple time-phased predictions (e.g., predictions at different future time points) rather than providing a single continuous trajectory. This segmentation allows the system to compute discrete predictions at key moments, reducing overall computational energy while eliminating track gapping by ensuring predictions exist at multiple time points that bridge gaps between observations.
Data Source
AI summary
A system and methodology/processes for non-kinematic/behavioral mapping to a local area abstraction (LAA) includes a technique for populating an LAA wherein human behavior or other non-strictly-kinematic motion may be present.