Dynamic Video Redaction Masks for Preserving Tracked Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic redaction systems face challenges in efficiently redacting video footage while preserving areas of interest without excessive computational overhead, particularly when objects move into overinclusive redaction masks, leading to unnecessary redaction of non-targeted regions.
Innovation Solution
A dynamic redaction method that switches between two techniques: using an overinclusive redaction mask for initial redaction and, upon overlap, modifies the mask to conform to the actual object boundaries, ensuring only necessary redaction occurs, thus balancing computational efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If an overinclusive redaction mask is used to simplify the rediction process, then the redaction speed increases and computational power required decreases, but areas that did not need to be redacted are obscured
Solution Approach 1:
The system dynamically adjusts the redaction mask based on the tracked region. When the tracked region intersects with the redaction mask, the system modifies the mask to un-redact portions that overlap with the tracked region, making the rediction process adaptive rather than static
Solution Approach 2:
The system applies different rediction treatments to different regions of the image. Areas covered by the rediction mask are redacted unless they intersect with tracked regions, where case the mask is modified to preserve those specific areas
2Measurement precision
If the actual border of the object is identified to prevent unnecessary rediction, then the accuracy of rediction improves, but the computational complexity increases and the process becomes slower
Solution Approach 1:
The system divides the image processing into two independent parts: object detection (identifying what to redact) and region tracking (identifying what to preserve). This segmentation allows each part to be optimized separately without requiring full boundary computation for all objects
Solution Approach 2:
The system performs tracking of regions of interest in advance and maintains this information throughout the video sequence. When rediction is applied, the pre-computed tracked regions are used to modify the mask, avoiding the need to compute boundaries during the rediction process
Data Source
AI summary
Techniques for dynamic redaction are provided. At least one object within a video to be redacted is identified. At least one region including the at least one object is redacted. The at least one redacted region including the at least one object and being larger than the at least one object. A tracked region being separate from the at least one object is identified. When the tracked region intersects with the at least one redacted region, the at least one redacted region is modified to un-redact portions of the at least one redacted region that overlap the tracked region.