Video Retrieval Apparatus Segmentation for Speed and Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing surveillance systems cannot continuously retrieve a person's location from past video images without decreasing processing speed, as they either fail to provide real-time updates for stray children or missing passengers due to slow processing speeds when retroactively retrieving images.
Innovation Solution
An information processing apparatus with an accepting unit for retrieval instructions, a first retrieving unit for processing past video images, and a second retrieving unit for processing future images, allowing continuous retrieval of a target person from both past and present video feeds without compromising processing speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If retroactive retrieval is performed on past video images to find a target person, then the ability to locate the person is improved, but the processing speed decreases
Solution Approach 1:
The patent divides the retrieval system into two separate retrieval units: a first retrieval unit that processes past video images stored in a database, and a second retrieval unit that processes current/future video images. This segmentation allows each unit to be optimized for its specific function, with the first unit handling historical data comprehensively and the second unit providing rapid real-time retrieval, thus resolving the contradiction between retrieval completeness and processing speed.
Solution Approach 2:
The system performs preliminary actions by continuously storing and indexing video images and their extracted features in a database before retrieval is needed. The first retrieval unit can immediately query this pre-prepared database without performing real-time processing, enabling fast retroactive retrieval of past images while maintaining high processing speed.
2Reliability
If real-time retrieval is performed on current video images to inform users of current whereabouts, then the ability to secure the person is improved, but the processing speed decreases
Solution Approach 1:
The patent separates real-time retrieval operations into a dedicated second retrieval unit that processes current video images independently from the first retrieval unit. This allows real-time retrieval to be performed on current frames without the computational overhead of comprehensive retroactive analysis, maintaining both real-time capability and processing speed.
Solution Approach 2:
The system extracts and stores feature data (copies) of persons from video images in advance in the database. When retrieval is needed, the retrieval units query these pre-computed features rather than analyzing raw video images in real-time, enabling fast real-time retrieval without compromising processing speed.
3Measurement precision
If comprehensive video image retrieval is performed covering both past and future time points, then the retrieval accuracy is improved, but the processing complexity increases
Solution Approach 1:
The patent divides the comprehensive retrieval task into two specialized retrieval units operating on different time domains. The first retrieval unit handles past images by querying the database, while the second retrieval unit handles current/future images by processing incoming video streams. This segmentation maintains high retrieval accuracy across all time points while reducing overall system complexity through functional specialization.
Solution Approach 2:
The retrieval system is designed with multi-functional capability through the two retrieval units that can operate independently or in conjunction. The first retrieval unit provides retroactive search functionality, while the second provides real-time monitoring functionality. Together they achieve comprehensive retrieval accuracy across all time points without requiring a single complex system to handle all functions simultaneously.
Data Source
AI summary
A first retriever performs, based on a person feature of a person extracted from a video image before an acceptance time point of a retrieval instruction of a target person and stored and a feature extracted from the target person related to the retrieval instruction, a first retrieving process of retrieving the target person from the video image stored, a second retriever performs, based on a feature of a person extracted from a video image after the acceptance time point and the feature of the target person extracted from a query image of the retrieval instruction, a second retrieving process of retrieving the target person from the video image input after the acceptance time point, and the first retriever performs the first retrieving process to the video image input during a period from the acceptance time point to preparation completion of the second retrieving process and second retrieving process start.