Object Identification Reliability in Multi-Camera Surveillance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video surveillance systems face challenges in accurately tracking human figures across multiple cameras due to variations in appearance caused by camera installation positions and orientations, leading to decreased identification accuracy and improper association of objects.
Innovation Solution
An information processing apparatus with a detecting unit, extracting unit, managing unit, and deriving unit that records feature amounts by orientation or region, calculates reliability, and associates them to improve object identification precision, using machine learning and classifiers like SVM for object detection and tracking, and Kalman Filter for motion estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature amounts are stored for each orientation to improve identification accuracy, then identification precision is improved, but system complexity increases due to managing multiple feature amounts for different orientations
Solution Approach 1:
The patent transforms the feature amount representation by introducing orientation parameters. Instead of storing a single feature amount, the system stores feature amounts parameterized by orientation angles, allowing the same object to be identified regardless of its rotational state in the image.
Solution Approach 2:
The patent segments the feature amount storage by dividing it into multiple orientation-specific feature amounts. Each orientation segment stores feature amounts for specific angular ranges, enabling precise matching while maintaining organized storage structure.
2Quantity of substance
If a single feature amount is used to represent all appearances, then storage efficiency is improved, but identification accuracy decreases due to appearance variations
Solution Approach 1:
The patent introduces orientation as a parameter to the feature amount representation. By parameterizing feature amounts with orientation information, the system can efficiently store multiple appearance variants without requiring completely separate storage for each possible appearance.
3Speed
If feature amounts are calculated from temporary orientation data, then processing speed is improved, but identification reliability decreases due to unstable and insufficient feature amounts
Solution Approach 1:
The patent performs preliminary processing by pre-calculating and storing feature amounts for multiple orientations in advance. This preparation work is done before actual identification occurs, so during runtime, the system can quickly retrieve and compare pre-computed feature amounts, maintaining both speed and reliability.
Data Source
AI summary
An information processing apparatus includes a detecting unit, an extracting unit, a managing unit, and a deriving unit. The detecting unit detects an object from an image. The extracting unit extracts a feature amount of the object detected by the detecting unit. The managing unit manages an element in the object detected by the detecting unit in association with the feature amount of the object in the element. The deriving unit derives a reliability indicative of a certainty for identifying an object with respect to the feature amount of the element.