ROI Tracking via Intermittent AI Detection and Motion Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies require significant computational resources and time to detect regions of interest (such as faces and license plates) in all video frames for real-time blind processing, making it difficult to implement on edge computers.
Innovation Solution
An image processing device and method that intermittently detect regions of interest using artificial intelligence in some frames and estimate and track these regions between frames using motion vectors, reducing the computational load and enabling real-time processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If artificial intelligence is used to detect regions of interest in all video frames for blind processing, then privacy protection accuracy is improved, but computational load and processing time increase significantly
Solution Approach 1:
The video processing is segmented into two distinct phases: detection frames where AI-based ROI detection is performed, and estimation frames where motion vector-based tracking is used. This segmentation allows the computationally intensive AI detection to be applied only periodically rather than to every frame, significantly reducing overall computational load while maintaining privacy protection accuracy through continuous tracking of detected regions across frames.
Solution Approach 2:
The system maintains continuous privacy protection across all video frames by combining periodic AI detection with continuous motion vector-based tracking. The useful action of ROI detection is sustained through the tracking unit that follows detected regions across estimation frames using motion vectors, ensuring no privacy gaps occur between periodic AI detection cycles.
2Measurement precision
If artificial intelligence is used to detect regions of interest in all video frames, then detection accuracy is improved, but processing speed decreases
Solution Approach 1:
The AI-based detection is performed periodically on selected detection frames rather than continuously on every frame. The tracking unit then interpolates ROI positions in intervening estimation frames using motion vectors. This periodic action maintains detection accuracy at key moments while dramatically improving processing speed by avoiding repeated AI computation on every frame.
Solution Approach 2:
The system creates motion vector copies of ROI positions from detection frames to estimate positions in estimation frames. Instead of re-detecting ROIs in every frame using AI, the system copies and adjusts position information using motion vectors, preserving detection accuracy while enabling rapid processing of all intermediate frames.
3Reliability
If AI-based detection is performed on every frame, then region of interest detection reliability is improved, but real-time processing capability is lost
Solution Approach 1:
Processing is segmented into detection frames (with AI for high reliability) and estimation frames (with motion tracking for speed). This segmentation enables the system to achieve real-time processing by using lightweight motion tracking for most frames while periodically applying reliable AI detection to maintain overall detection reliability across the video sequence.
Solution Approach 2:
Motion vectors serve as an intermediary mechanism that bridges detection frames and estimation frames. The motion tracking unit uses these vectors to reliably estimate ROI positions in estimation frames without requiring full AI processing, thereby maintaining detection reliability while enabling real-time processing speed.
4Reliability
If AI processing is applied to all frames for blind processing, then privacy protection effectiveness is improved, but implementation on edge computers becomes difficult
Solution Approach 1:
The system applies AI processing partially only to detection frames rather than excessively to every frame. This partial action reduces computational requirements to levels suitable for edge computers with limited resources, while the motion tracking component ensures privacy protection effectiveness is maintained across all frames through continuous monitoring of detected regions.
Solution Approach 2:
The system substitutes the mechanical AI processing system with a lighter motion vector-based tracking system for estimation frames. This substitution replaces computationally intensive AI operations with simpler mathematical calculations, enabling implementation on resource-constrained edge computers while maintaining privacy protection effectiveness through continuous tracking.
Data Source
AI summary
The present disclosure provides an image processing device and method using region of interest detection and tracking, wherein the image processing device using region of interest detection and tracking, including an interest region detection unit that detects at least one region of interest (ROI) from multiple detection frames of a video using artificial intelligence; an interest region motion vector estimation unit that estimates an inter-frame interest region motion vector from at least one estimation frame located between neighboring detection frames having the detected region of interest; and an interest region tracking unit that tracks the region of interest in the estimation frame using the interest region motion vector. According to the present disclosure, the region of interest is detected using artificial intelligence intermittently rather than in each of all frames, thereby drastically reducing the amount of computation, and enabling real-time image blind processing and enabling implementation in an edge computer.