Video Privacy Masking via Periodic Detection and Motion Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video data processing technologies for surveillance cameras consume significant hardware resources and data throughput to detect and mask privacy-sensitive areas, such as human faces, in real-time video streams, and they struggle to efficiently identify masked objects on monitoring servers.
Innovation Solution
A method that involves storing video data, detecting objects of interest across multiple frames, estimating object positions using difference vectors, and performing masking processing to reduce data throughput while enabling easy identification of masked objects by encoding feature information alongside the video stream.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If object detection and masking is performed for each video frame, then privacy protection is achieved, but hardware resource consumption and data throughput requirements increase significantly
Solution Approach 1:
The system performs object detection only on selected key frames rather than every frame, and uses motion estimation to track objects in intermediate frames. This preliminary detection approach reduces computational load while maintaining privacy protection effectiveness across the entire video sequence.
Solution Approach 2:
The system adopts periodic object detection at intervals (e.g., every N frames) combined with continuous motion estimation. This periodic detection strategy maintains privacy protection while significantly reducing hardware resource consumption compared to frame-by-frame detection.
2Reliability
If object detection and masking is performed for each video frame, then privacy protection is achieved, but data throughput requirements increase significantly
Solution Approach 1:
The system performs object detection only on selected key frames rather than every frame, and uses motion estimation to track objects in intermediate frames. This preliminary detection approach reduces computational load while maintaining privacy protection effectiveness across the entire video sequence.
Solution Approach 2:
The system adopts periodic object detection at intervals (e.g., every N frames) combined with continuous motion estimation. This periodic detection strategy maintains privacy protection while significantly reducing hardware resource consumption compared to frame-by-frame detection.
3Reliability
If masking processing is applied to protect privacy, then privacy-sensitive information is obscured, but identification of masked objects becomes difficult
Solution Approach 1:
The system extracts and transmits metadata containing object identification information (such as bounding box coordinates, object types, and detection confidence) separately from the masked video stream. This allows monitoring servers to identify objects of interest without requiring unmasking, thus maintaining both privacy protection and object identification capability.
Solution Approach 2:
The system introduces metadata as an intermediary carrier that conveys object identification information without revealing the actual visual content. This metadata acts as a mediator between the masked video stream and the monitoring server, enabling object identification while preserving privacy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided is a method of processing video data, which includes: storing video data captured by a camera; detecting an object of interest from a plurality of video frames of the stored video data; performing masking processing on an object area of interest including the detected object of interest; and encoding the video data which is subjected to the masking processing to generate a video stream, wherein the masking processing is performed by estimating a position of the object of interest in a video frame located between two or more video frames for which a difference vector of the object of interest has been calculated, based on the difference vector of the object of interest detected in the two or more video frames..