Video Redaction via Probability-Based Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video redaction technologies for surveillance and body-worn cameras are inefficient and costly due to high false negatives and the need for manual frame-by-frame review, which is time-consuming and labor-intensive, especially in complying with privacy laws like the Freedom of Information Act.
Innovation Solution
A computer-implemented method and system that uses probability-based analysis for object detection and redaction, incorporating a video acquisition module, object detection module, redaction module, and manual review module to generate and review redacted video, with adaptive frame rates based on object detection probabilities, enhancing automation and reducing manual review time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated object detection and redaction is used, then redaction speed is improved, but false negatives increase
Solution Approach 1:
The patent introduces probability thresholds as an intermediary mechanism between automated detection and manual review. Objects with detection probability below the threshold are flagged for manual review, while those above the threshold are automatically redacted. This intermediary threshold resolves the contradiction by enabling high-speed automated processing for high-confidence detections while maintaining accuracy through selective manual verification of uncertain cases.
Solution Approach 2:
The patent applies different processing quality levels to different objects based on their detection probability. High-probability objects receive automated redaction (high speed, acceptable accuracy), while low-probability objects receive manual review (lower speed, higher accuracy). This local differentiation of processing quality allows the system to optimize both speed and accuracy simultaneously.
2Reliability
If manual frame-by-frame review is performed, then redaction accuracy is improved, but time consumption increases
Solution Approach 1:
The patent segments the video review process into two distinct segments: automated redaction of high-probability objects and manual review of low-probability objects. This segmentation allows reviewers to focus only on the small portion of video requiring human attention, dramatically reducing review time while maintaining accuracy for uncertain cases.
Solution Approach 2:
The automated detection system performs preliminary action by pre-screening all video frames and identifying objects with high detection probability before they reach manual review. This preliminary automated processing eliminates the need for reviewers to examine every frame, reducing time consumption while ensuring accuracy through targeted human review of only the necessary frames.
3Measurement precision
If probability threshold is set high, then false positives are reduced, but false negatives increase
Solution Approach 1:
The patent implements feedback through manual review of low-probability detections. Reviewers provide feedback on objects flagged by automated detection, allowing the system to learn from mistakes and adjust thresholds. This feedback loop enables the system to maintain high precision by reviewing uncertain cases while improving overall reliability through continuous learning from reviewer corrections.
Data Source
AI summary
Disclosed is a method and system for generating redacted video of a scene captured using a video camera and augmenting a manual review process of the redacted video of the scene. According to an exemplary embodiment, a video of the scene is redacted via probability-based analysis to detect and obscure privacy sensitive objects included in the captured video. A manual review process of the generated redacted video is augmented to use the object detection probability to enhance the reviewing video frame rate to expedite the manual review process.