Intersection Pedestrian Behavior Prediction for Real-Time Hazard Warning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting pedestrian crossing behavior at intersections require complex mathematical models or large labeled data sets, failing to account for interdependence and randomness in pedestrian behaviors influenced by factors like age, gender, and psychology, and are inadequate for real-time hazard prediction.
Innovation Solution
A method using millimeter wave radar and visual cameras to detect pedestrian behavior, employing a fully convolutional neural network-long-short term memory network (FCN-LSTM) model trained via reinforcement learning to predict pedestrian actions like walking, stopping, and running, without requiring pre-established models or extensive data sets, and providing hazard warnings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If model-based methods (social force model, energy function model, Markov model) are used to predict pedestrian behavior, then the pedestrian movement can be described through mathematical expressions, but it is not possible to construct a clear mathematical model to describe the pedestrian crossing behavior at the intersection due to individual differences and randomness
Solution Approach 1:
The patent replaces complex mathematical modeling (mechanical/systematic approach) with deep learning-based feature extraction (data-driven approach). Instead of constructing explicit mathematical models of pedestrian movement forces and interactions, the system uses neural networks to automatically learn behavior patterns from video data, substituting the mechanical modeling process with an intelligent data processing system that handles individual differences and randomness more effectively
Solution Approach 2:
The patent changes the fundamental parameters of the prediction system by transitioning from fixed mathematical model parameters to dynamic learned features. The system extracts behavior features (walking speed, direction, stopping probability, running probability) that are continuously updated based on observed pedestrian actions, allowing the model to adapt to individual differences and random behaviors without requiring complex theoretical formulations
2Productivity
If data-driven deep learning methods (RNN, LSTM, S-LSTM, GAN, GAT) are used to predict pedestrian behavior, then continuous behavior can be regarded as time series for serialized prediction, but it is difficult to obtain massive labeled data sets to extract interdependence and randomness features
Solution Approach 1:
The patent applies preliminary action by pre-defining specific behavior features (walking speed, direction, stopping probability, running probability) that are relevant to intersection crossing scenarios. Instead of requiring massive labeled datasets for general pedestrian behavior, the system pre-identifies the key features needed for intersection safety prediction, then trains the model to extract these specific features from video data, reducing the data requirement while maintaining prediction capability
Solution Approach 2:
The patent applies local quality by focusing on specific behavior features rather than attempting to model all aspects of pedestrian behavior. The system extracts localized behavior characteristics (walking speed, direction, stopping probability, running probability) that are most relevant to intersection crossing, rather than requiring comprehensive data about all pedestrian attributes and behaviors. This selective feature extraction reduces data requirements while improving prediction relevance
3Measurement precision
If complex mathematical models or large labeled data sets are required for pedestrian behavior prediction, then prediction accuracy can be improved, but the system becomes inadequate for real-time hazard prediction due to computational complexity
Solution Approach 1:
The patent extracts only the essential behavior features needed for real-time intersection hazard prediction: walking speed, direction, stopping probability, and running probability. By extracting these specific features rather than processing complete pedestrian behavior data, the system achieves accurate prediction while reducing computational complexity and enabling real-time processing. The extraction focuses on features that directly impact collision risk assessment
Data Source
AI summary
A method for predicting a pedestrian crossing behavior for an intersection includes the following steps: step 1: designing an immediate reward function; step 2: establishing a fully convolutional neural network-long-short term memory network (FCN-LSTM) model to predict a motion reward function; step 3: training the fully convolutional neural network-long-short term memory network (FCN-LSTM) model based on reinforcement learning; and step 4: predicting the pedestrian crossing behavior and performing hazard early-warning. The technical solution does not require establishment of a complex pedestrian movement model or preparation of massive labeled data sets, achieves autonomous learning of pedestrian crossing behavior features at the intersection, predicts their walking, stopping, running and other behaviors, especially predicts the pedestrian crossing behavior when inducing hazards such as pedestrian-vehicle collision and scratch in real time, and performs hazard early-warning on crossing pedestrians and passing vehicles.