Machine Learning Engine for Video Behavior Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video surveillance systems require pre-defined knowledge of behaviors to recognize abnormal activities, making them labor-intensive and costly, as they cannot accurately characterize behaviors without prior programming and are limited to detecting predefined types of behavior.
Innovation Solution
A machine-learning engine that analyzes video frames to recognize and distinguish between normal and abnormal behaviors by generating semantic and phase-space symbol streams, combining them to form vector representations, and using a cognitive model to identify patterns and generate alerts for unusual patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pre-defined behavior codes are used to recognize abnormal activities, then the system can detect specific predefined behaviors, but the system becomes labor-intensive and costly due to extensive programming requirements
Solution Approach 1:
The system performs self-learning by automatically analyzing video data to discover and classify behaviors without requiring manual programming. The machine learning engine autonomously builds behavior models from observed data, eliminating the need for extensive pre-programming while maintaining reliable behavior detection
Solution Approach 2:
The system transforms behavior recognition from a static, pre-programmed approach to a dynamic, data-driven approach. By changing the parameter of behavior definition from fixed codes to learned patterns, the system achieves both reliability in detection and reduction in programming complexity
2Reliability
If separate software products are developed for distinct behaviors, then each behavior can be accurately recognized, but the system becomes prohibitively costly and labor-intensive
Solution Approach 1:
A single machine learning engine is designed to handle multiple different behaviors simultaneously. The system universally processes various behavior types (e.g., lurking, swimming patterns, fighting) through one unified platform, eliminating the need for separate software products while maintaining accurate recognition for each behavior type
Solution Approach 2:
The system transitions from static, behavior-specific software to a dynamic, adaptive machine learning model that can automatically adjust to recognize multiple behavior types. The learned behavior models are updated continuously based on new data, enabling the single system to efficiently handle diverse behaviors
3Ease of manufacture
If surveillance systems are pre-programmed to detect limited ranges of behavior, then development time is reduced, but the system cannot accurately characterize or detect behaviors outside the predefined range
Solution Approach 1:
The system performs preliminary learning during a deployment phase where it automatically analyzes video data to discover behaviors in the specific environment. This preliminary action of self-learning enables the system to adapt to the specific context before full operation, achieving both quick deployment and broad behavior detection capability
Solution Approach 2:
The system continuously learns from observed behaviors and updates its models based on feedback from the environment. This feedback mechanism allows the system to expand its behavior detection range beyond initial programming by automatically adapting to new behavior patterns it encounters during operation
Data Source
AI summary
A machine-learning engine is disclosed that is configured to recognize and learn behaviors, as well as to identify and distinguish between normal and abnormal behavior within a scene, by analyzing movements and/or activities (or absence of such) over time. The machine-learning engine may be configured to evaluate a sequence of primitive events and associated kinematic data generated for an object depicted in a sequence of video frames and a related vector representation. The vector representation is generated from a primitive event symbol stream and a phase space symbol stream, and the streams describe actions of the objects depicted in the sequence of video frames.