Human Flow Analysis Grouping for Occlusion Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human flow prediction systems struggle to accurately predict the movement of individuals when they temporarily disappear from camera view due to occlusion or other causes, leading to inaccurate line of movement prediction.
Innovation Solution
A human flow analysis method that acquires movement information within a predetermined space, extracts associated individuals based on proximity and positional relationships, identifies association information through facial and gaze analysis, and predicts behavior by grouping individuals together to enhance prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If individual movement prediction is performed for each person, then the system complexity is reduced, but the prediction accuracy deteriorates when persons are temporarily occluded
Solution Approach 1:
The patent merges multiple individual prediction models into a group-based prediction model. When persons are detected to be moving together (within a predetermined distance threshold), their movement predictions are combined into a single group prediction, allowing the system to maintain accuracy even when individual persons are occluded, while avoiding the complexity of tracking each person separately throughout the group
2Measurement precision
If association analysis between persons is performed, then the prediction accuracy is improved, but the device complexity increases
Solution Approach 1:
The patent segments the population into individual persons and groups based on spatial proximity and association rules. By dividing the prediction task into individual-level analysis (for determining associations) and group-level prediction (for movement forecasting), the system achieves high accuracy without requiring a completely complex unified model
Solution Approach 2:
The patent implements dynamic grouping where persons are continuously evaluated against association rules and distance thresholds. Groups are formed, maintained, or dissolved based on real-time movement patterns, allowing the system to adapt its complexity level dynamically - using group prediction only when associations are detected, rather than always applying complex group analysis
Data Source
AI summary
A human flow analysis apparatus includes a movement information acquirer that acquires movement information, the movement information representing a history of movement within a predetermined space by multiple persons moving within the predetermined space, an associated-nodes extractor that, based on the movement information, extracts at least two persons assumed to be moving in association with each other, an association information identifier that identifies association information, the association information indicating what association the extracted at least two persons have with each other, a node fusion determiner that, based on the identified association information, determines whether to group the at least two persons together, and a behavior predictor that predicts a behavior of the at least two persons who have been determined to be grouped together.