Feature Amount Registration Timing for Face Collation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing apparatuses face delays in registering feature amounts from human bodies or parts, leading to collation failures due to unregistered faces with preferable directions, sizes, and eye capturing conditions.
Innovation Solution
An information processing apparatus that acquires video, detects and tracks human bodies or parts, and registers feature amounts during a specified tracking period, including a timing-based registration approach to reduce delay and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all faces detected from all frames of a video are registered in a database, then collation accuracy is improved, but the database becomes bulky and processing time increases
Solution Approach 1:
The patent applies preliminary action by registering feature amounts during a first period from the timing of start of tracking to a timing before end of tracking. This advance registration ensures that feature amounts are available in the database before they are needed for collation, reducing delay time while controlling the quantity of stored data through selective time-window registration.
2Productivity
If frames are thinned out to lower the frame rate of processing, then processing speed is improved, but feature amounts of faces with preferable directions, sizes, and eye capturing conditions may be unregistered
Solution Approach 1:
The system performs preliminary registration of feature amounts during the tracking period before collation is needed. This allows the system to process frames at lower rates while ensuring that high-quality feature amounts (with preferable directions, sizes, and eye capturing conditions) are captured and stored in advance, eliminating the need to process every frame at high speed.
Solution Approach 2:
The tracking unit automatically identifies and tracks faces with preferable characteristics (directions, sizes, eye capturing conditions) and triggers feature amount registration for these specific cases. This self-service mechanism ensures that only high-quality feature amounts are registered, maintaining collation accuracy while reducing the overall number of registrations compared to processing all frames.
3Quantity of substance
If feature amounts are registered only at the end of tracking, then database size is reduced, but delay time from detection to registration increases
Solution Approach 1:
The patent implements preliminary action by registering feature amounts during a first period from the timing of start of tracking to a timing before end of tracking. This advance registration reduces the delay time from detection to registration availability, ensuring that feature amounts are in the database when needed for collation while still controlling database size through selective time-window registration.
Data Source
AI summary
This invention prevents a collation database from becoming bulky and shortening the delay time needed from person detection to registration in the collation database. An information processing apparatus comprises an acquiring unit which acquires a video, a detecting unit which detects a whole body or a part of a person from at least one frame of the acquired video, a tracking unit which tracks the whole body or the part of the person detected, and a registering unit which registers, in a database, a feature amount extracted from the whole body or the part of the person tracked during a first period from a timing of a start of tracking of the whole body or the part of the person by the tracking unit to a timing before an end of the tracking by the tracking unit.