Abnormal Movement Detection Using Online Topic Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing movement learning methods in images are limited by their batch learning schemes, which fail to reflect continuous variations in images, especially in situations with time differences in normal movement patterns and complex scenarios like crowd concentrations.
Innovation Solution
An apparatus and method utilizing a feature tracing unit, topic online learning unit, and movement pattern online learning unit that employs online learning techniques, such as probabilistic topic models and K-means clustering, to extract features, classify trajectories, and infer spatiotemporal correlations, enabling the detection of abnormal movements even in varying and complex images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If batch learning scheme is used to learn movement patterns, then learning accuracy for static patterns is improved, but the system cannot reflect continuous variations in images over time
Solution Approach 1:
The patent transitions from static batch learning to dynamic online learning by continuously updating the movement pattern model as new image data arrives. The system dynamically adapts to changing movement patterns in real-time, allowing the learning model to evolve with continuous variations in the monitored environment while maintaining accurate detection of abnormal movements.
2Productivity
If trajectory-based learning method is used, then movement patterns can be clustered into main patterns, but the method fails to handle arbitrary angle projections and cut-off trajectories effectively
Solution Approach 1:
The patent transforms trajectory data into a probabilistic representation using Gaussian mixture models, changing the parameter space from direct trajectory coordinates to probability distributions. This transformation allows the system to handle arbitrary angle projections and partial trajectories by modeling them as probabilistic patterns rather than requiring complete, aligned trajectory data.
3Loss of information
If local feature-based learning with optical flow is used, then velocity and direction information can be extracted, but the method cannot handle complex scenarios like crowd concentration images
Solution Approach 1:
The patent merges multiple learning approaches by combining trajectory-based methods with local feature-based optical flow extraction. By integrating these methods within a unified probabilistic framework using Gaussian mixture models and online learning, the system can handle complex scenarios like crowd concentrations while preserving velocity and direction information extraction capabilities.
Data Source
AI summary
Provided are a method and apparatus for detecting an abnormal movement. The apparatus includes a feature tracing unit configured to extract features of a moving object in an input image, trace a variation in position of the extracted features according to time, and ascertain trajectories of the extracted features; a topic online learning unit configured to classify the input image in units of documents which are bundles of the trajectories, and ascertain probability distribution states of topics, which constitute the classified document, by using an online learning method which is a probabilistic topic model; and a movement pattern online learning unit configured to learn a velocity and a direction for each of the ascertained topics, and learn a movement pattern by inferring a spatiotemporal correlation between the ascertained topics.