Cross-camera object detection via background-relative indexing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Variations in illumination conditions and camera models lead to inaccuracies in identifying objects across multiple videos captured by different cameras, affecting the accuracy of object recognition.

Innovation Solution

An image processing apparatus that extracts attribute values of image regions, including background objects and objects, and calculates a relative index to detect the same object across videos by referencing the attribute values of background objects, thereby reducing the impact of varying conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If attribute information is extracted and compared directly from videos captured by multiple cameras, then object identification can be implemented across cameras, but accuracy deteriorates due to variations in illumination conditions and camera models

Engineering Contradiction:

Improveobject identification across camerasVSAvoididentification accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent transforms absolute attribute values (which vary with illumination and camera model) into relative index values by comparing each object's attributes against background object attributes captured at the same time. This parameter transformation eliminates the influence of environmental variations while preserving object identification capability across different cameras and conditions

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

Background objects serve as an intermediary reference standard. By comparing target objects against background objects captured simultaneously, the system creates a relative reference frame that mediates between different camera characteristics and illumination conditions, enabling accurate cross-camera identification

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If attribute information is extracted from both background objects and target objects, then the influence of varying conditions can be reduced, but the processing complexity increases

Engineering Contradiction:

Improveidentification accuracyVSAvoidprocessing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the image into background object regions and target object regions, processing them separately. Background objects are used to establish environmental reference values, while target objects are identified by comparing their attributes against these references. This segmentation simplifies the overall processing by dividing it into manageable, independent steps

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS10762372B2Image processing apparatus and control method therefor

Publication Date: 2020.09.01 CANON KK

AI summary

An image processing apparatus inputs a first and a second captured videos; extracts attribute values of image regions of background objects commonly included in the first and the second captured videos respectively; extracts attribute values of image regions of objects respectively included in the first and the second captured videos; derives a relative index of an attribute value of the image region of the object with reference to the attribute value of the image region of the background object with respect to the object included in each of the first and the second captured videos; and detects the same object included in the first and the second captured videos based on the index derived with respect to the object included in each of the first and the second captured videos.