Grid-Map Motion Prediction for High-Accuracy Multi-Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting the future behavior of moving objects in complex environments face challenges in balancing prediction accuracy with the increasing calculation load, often sacrificing accuracy to reduce computational demands.
Innovation Solution
A moving object behavior prediction device that generates a single input map dividing the environment into cells, using a convolutional neural network to estimate feature amounts and predict future positions, thereby sharing information across multiple objects to suppress calculation increases while improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a wide range of environmental information is arithmetically processed for each moving object, then the prediction accuracy of future behavior is improved, but the calculation amount increases explosively
Solution Approach 1:
The environment is divided into a grid map with multiple cells, where each cell stores related information of static objects and moving objects. This segmentation allows the system to process environmental information in a structured manner, reducing the computational complexity of analyzing wide-range environmental data for each moving object while maintaining prediction accuracy.
Solution Approach 2:
The patent merges the processing of environmental information by using a single grid map that contains information about all moving objects and static objects in the environment. Instead of processing environmental data separately for each moving object, the system performs arithmetic processing on the unified grid map, which suppresses the explosive increase in calculation amount while still enabling accurate prediction of each moving object's future behavior through shared environmental context.
2Productivity
If only local environmental information is arithmetically processed to reduce calculation amount, then the calculation load is manageable, but the prediction accuracy of future behavior is deteriorated
Solution Approach 1:
The grid map serves as a universal data structure that contains environmental information applicable to all moving objects in the scene. By performing arithmetic processing on this unified grid map, the system extracts environmental features that are relevant to multiple moving objects simultaneously. This universal processing approach reduces calculation load compared to individual processing for each object, while still providing accurate prediction capabilities through the shared environmental context captured in the grid map.
Data Source
AI summary
A moving object behavior prediction device improves prediction accuracy and includes: an input map generation unit that generates a single input map in which a region capable of containing a plurality of the moving objects is divided into a plurality of cells, in which each of the cells stores related information of a static object and related information of the moving object; a movement amount estimation unit that estimates a movement amount as a feature amount of each cell from the input map, using a trained convolutional neural network; a movement amount acquisition unit that acquires a movement amount at a current position of the moving object from a movement amount stored in a peripheral cell the moving object; and a future position prediction unit that predicts a future position of the moving object based on a feature amount at a current position of the moving object.