Video Analysis Long-Term Memory for Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video surveillance systems are labor-intensive and costly to maintain, as they require pre-defined object and behavior recognition, limiting their ability to detect new patterns or changes in existing patterns, and are often ineffective in associating related aspects of behavior.
Innovation Solution
A video analysis system with a long-term memory that learns and detects anomalous events by using a semantic symbol stream to query a neural network, combining episodic and long-term memory to determine the occurrence of anomalous events through distance measurement based on cosine similarity and occurrence frequency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pre-defined object and behavior recognition patterns are used in video surveillance systems, then the system can recognize specified objects and behaviors, but the system becomes labor-intensive and costly to maintain when new patterns need to be recognized
Solution Approach 1:
The system automatically learns and updates behavior patterns through machine learning algorithms, eliminating the need for manual software development and configuration when new patterns need to be recognized. The system serves itself by continuously improving its recognition capabilities through automated learning from video data.
Solution Approach 2:
The system changes its recognition parameters dynamically by learning from observed behaviors and adapting its patterns accordingly. Instead of using fixed pre-defined patterns, the system modifies its recognition parameters based on learned behaviors, allowing it to recognize new patterns without manual intervention.
2Reliability
If pre-defined behavior patterns are used for surveillance, then the system can detect specified behaviors, but the system is incapable of recognizing new patterns of behavior that may emerge
Solution Approach 1:
The system transitions from static pre-defined patterns to dynamic learned patterns that adapt over time. The behavior patterns are not fixed but evolve as the system learns from observed behaviors, enabling the system to recognize new patterns while maintaining reliability in detecting known patterns.
Solution Approach 2:
The system uses feedback from observed behaviors to continuously improve and update its recognition patterns. By analyzing detected behaviors and feeding this information back into the learning system, the system adapts to new patterns while maintaining reliable detection of established patterns.
3Measurement precision
If static recognition patterns are used in video surveillance, then the system can identify predefined behaviors, but the system produces false positives and is either under inclusive or over inclusive
Solution Approach 1:
The system uses multiple overlapping behavior patterns with varying specificity levels. Instead of relying on a single static pattern, the system employs multiple patterns that collectively cover a broader range of behaviors, reducing false positives while maintaining detection precision through the combined evaluation of multiple partial matches.
Data Source
AI summary
Techniques are described for detecting anomalous events using a long-term memory in a video analysis system. The long-term memory may be used to store and retrieve information learned while a video analysis system observes a stream of video frames depicting a given scene. Further, the long-term memory may be configured to detect the occurrence of anomalous events, relative to observations of other events that have occurred in the scene over time. A distance measure may used to determine a distance between an active percept (encoding an observed event depicted in the stream of video frames) and a retrieved percept (encoding a memory of previously observed events in the long-term memory). If the distance exceeds a specified threshold, the long-term memory may publish the occurrence of an anomalous event for review by users of the system.