Image Recognition Method for Surveillance Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional surveillance devices require high labor costs due to the need for multiple devices to be monitored by security guards, and they consume significant power for image recognition processing, which is a challenge especially in off-the-grid areas where power supply is limited.
Innovation Solution
An image recognition method that analyzes pixel value distributions across multiple images to generate recognition signals, reducing the need for continuous power consumption by processing images locally and only transmitting relevant data, using a processor connected to a camera, storage medium, and signal transceiver, with adjustable thresholds to minimize false alerts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image recognition is performed at the surveillance device end, then the ability to detect abnormalities is improved, but power consumption increases significantly
Solution Approach 1:
The image recognition process is segmented into multiple stages: first extracting only changed regions (differences from reference images), then performing recognition only on these changed portions. This segmentation allows the system to maintain high detection accuracy while significantly reducing the computational load and power consumption compared to processing entire images.
Solution Approach 2:
The system performs partial image recognition by focusing only on the changed regions rather than the entire image. By applying recognition algorithms only to the differential portions where changes occur, the system achieves effective abnormality detection with reduced computational resources and lower power consumption.
2Reliability
If every 9 to 16 surveillance devices need a security guard for monitoring, then abnormal events can be detected, but labor cost increases
Solution Approach 1:
The surveillance devices perform self-monitoring through automated image recognition algorithms. Each device compares captured images against reference images, automatically detects changes, and generates alerts without human intervention. This self-service capability eliminates the need for security guards to manually monitor multiple devices, thereby reducing labor costs while maintaining reliable abnormal event detection.
Solution Approach 2:
The system implements automated feedback loops where image recognition results directly trigger alerts or notifications. When changes are detected through comparison with reference images, the system automatically generates warnings without requiring human review, enabling reliable monitoring with minimal human resources.
3Use of energy by moving object
If captured image streams are sent to the control center for image recognition, then processing power requirements at the device end are reduced, but network connectivity and continuous power supply are required
Solution Approach 1:
The system performs preliminary image processing and change detection at the device end before any potential transmission. By pre-processing images locally to identify only meaningful changes, the device reduces dependency on continuous network connectivity and can operate autonomously in off-the-grid scenarios, enhancing adaptability while maintaining low power consumption.
Data Source
AI summary
Disclosed is an image recognition method including: producing a distribution of first pixel value range by acquiring a distribution of pixel values of a plurality of pixels of a first selected block in a first surveillance image from previous M images; producing a distribution of second pixel value range by acquiring a distribution of pixel values of the pixels of the first selected block from previous N images, wherein N and M are positive integers, and N<M; obtaining a first varying parameter related to the first selected block according to the distribution of first pixel value range and the distribution of second pixel value range; and generating a first recognition signal when the first varying parameter is greater than a first threshold.