Trajectory Clustering for Behavior Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current behavior recognition techniques in surveillance systems for public places lack accuracy in calculating the likelihood of trajectory data representation by model data, which is crucial for distinguishing normal from anomalous behavior.
Innovation Solution
An information processing apparatus and method that acquires trajectory data, clusters it into groups based on probabilistic distribution, generates representative velocity and latent position distributions, computes scaling factors, and updates group identity distributions to create model data for accurate behavior classification, determining normal or abnormal behavior based on likelihood calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If trajectory data is divided into groups based on velocity profiles, then behavior modeling is simplified, but the accuracy of likelihood calculation deteriorates
Solution Approach 1:
The patent segments trajectory data into multiple groups based on velocity profiles, where each group contains trajectories with similar motion characteristics. This segmentation allows the system to model different behaviors separately, simplifying the overall modeling complexity while maintaining accuracy through group-specific velocity profiles and latent position distributions.
Solution Approach 2:
The patent introduces a latent dimension by computing latent position distributions for each trajectory, which captures unobserved motion patterns. This additional dimensional representation allows the system to accurately calculate likelihoods even when trajectories are grouped, resolving the contradiction between simplified grouping and accurate likelihood calculation.
2Measurement precision
If group identity distribution is updated based on likelihood, then behavior recognition accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent implements a feedback mechanism where the group identity distribution is updated based on the calculated likelihoods. This feedback loop allows the system to iteratively refine group assignments and velocity profiles, improving recognition accuracy. The feedback process is computationally intensive but necessary for achieving accurate behavior modeling.
Solution Approach 2:
The system performs self-service by automatically updating group identities and velocity profiles without external intervention. The likelihood calculations and group reassignments occur autonomously based on the input trajectory data, reducing the need for manual configuration while maintaining high accuracy.
Data Source
AI summary
The information processing apparatus (2000) of the example embodiment 1 includes an acquisition unit (2020), a clustering unit (2040), and a modeling unit (2060). The acquisition unit (2020) acquires a plurality of trajectory data. Until a predetermined termination condition is satisfied, the clustering unit (2040) repeatedly performs: 1) dividing the plurality of trajectory data into one or more groups using a group identity distribution of each trajectory data; 2) determining a time-sequence of representative velocity for each group; 3) determining, for each trajectory data, a time-sequence of a latent position distribution of a corresponding object; and 4) determining a scaling factor for each trajectory data; and 5) updating the group identity distribution of each trajectory data. The modeling unit (2060) generates a model data for each group. The model data includes the time-sequence of representative velocity generated by the clustering unit (2040).