Movement Mode Learning With Map-Aware Pedestrian Trajectories
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing model parameter learning methods for movement mode models do not effectively consider stationary objects when learning movement trajectories, leading to potential interference with both moving and stationary objects in environments with multiple objects.
Innovation Solution
A method that synchronizes time series data of movement mode parameters, positional relationships, and environment information to learn model parameters for movement mode models, using this data to determine movement mode parameters that avoid interference with both moving and stationary objects by incorporating two-dimensional map information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If mask image is created using movement trajectory of reference pedestrian and nearby pedestrian, then movement direction can be determined, but arrangement of stationary object such as wall is not considered in learning
Solution Approach 1:
The patent combines multiple data sources (movement trajectory data, environment information including stationary objects, and map information) into a unified learning dataset. This merging allows the movement mode model to simultaneously consider both moving objects (pedestrians) and stationary objects (walls, structures) when determining movement direction, resolving the contradiction between ease of operation and reliability.
Solution Approach 2:
The patent performs preliminary data preparation by acquiring and processing environment information about stationary objects before creating the mask image for learning. This preliminary action ensures that stationary object arrangement is incorporated into the learning data structure, allowing the model to consider both moving and stationary objects during movement determination without compromising operational simplicity.
2Adaptability or versatility
If environment information is acquired as two-dimensional map information, then the two-dimensional map can be used when determining movement mode parameter, but additional data processing is required
Solution Approach 1:
The patent makes the movement mode model adaptable to different environment representations by accepting two-dimensional map information as input. The model is designed to process various types of environment information (including but not limited to 2D maps) in a unified manner, allowing versatile application across different scenarios while managing processing complexity through standardized data structures.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In a learning method, a time series of a movement direction when a reference pedestrian M1 moves to a destination multiple times, a time series of a mask image indicating a positional relationship of nearby pedestrians M2 in a movement direction of the reference pedestrian M1, and a time series of an environment information image 35 are acquired, learning data is created by associating these time series with each other, and a model parameter of a CNN 33a is learned by a back propagation method using the learning data.