Video Redaction Bounding Regions for Accurate Privacy Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video surveillance systems face challenges in efficiently and accurately redacting identifying images of individuals to comply with privacy regulations, as manual redaction is time-consuming and imprecise.
Innovation Solution
A system and method for automated video redaction that uses object detection and tracking to identify and replace pixels within bounding regions of individuals in video frames, allowing user selection and ensuring accurate anonymization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual redaction is used, then accuracy can be maintained, but time consumption increases significantly
Solution Approach 1:
The patent replaces manual mechanical redaction processes with an automated computer vision system that uses object detection algorithms (such as YOLO or SSD models) to automatically identify and redact individuals in surveillance footage, eliminating the need for manual frame-by-frame processing while maintaining accuracy through algorithmic person detection and tracking
Solution Approach 2:
The system performs self-service by automatically detecting, tracking, and redacting individuals without requiring continuous human intervention. The automated pipeline processes entire video streams independently, applying redaction masks to detected persons across multiple frames without manual operation
2Productivity
If automated redaction is used, then time consumption decreases, but redaction precision may deteriorate
Solution Approach 1:
The patent implements feedback mechanisms where detected persons are tracked across multiple video frames using tracking algorithms (such as Kalman filtering or deep SORT), and the system continuously refines bounding box positions and redaction mask applications based on detected motion patterns and spatial consistency, ensuring high precision through iterative verification
Solution Approach 2:
The system segments the redaction task into distinct modular components: person detection, person tracking, bounding box refinement, and redaction mask application. Each component operates independently with specialized algorithms, allowing the system to maintain high precision through optimized processing of each segment while achieving overall automation
3Object-affected harmful factors
If all people in video are redacted, then privacy protection increases, but information loss increases
Solution Approach 1:
The patent applies local quality by selectively redacting only the specific regions where individuals are detected, rather than redacting entire frames or all content. The redaction masks are precisely applied to bounding boxes surrounding detected persons, leaving the rest of the surveillance footage intact and preserving useful information about events, locations, and other non-person elements
Data Source
AI summary
Aspects of the embodiments described herein are related to systems, methods, and computer products for forming a redacted video. A video is processed to detect one or more subjects present within the video. From the detected subjects, a subject to be redacted is determined. Bounding regions are determined for the subject in each frame in which the subject is present. Determining the bounding regions includes calculating uncertainty values related to the level of uncertainty regarding the subject's position and adjusting the size of the bounding regions based on the uncertainty values. A redacted video is then formed by redacting the subject from the video using the bounding regions.