Person Detection Using Angle-Matched Feature Selection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The specific person detection system faces accuracy degradation due to direction fluctuation of the person not being aware of the camera, leading to overlooking and erroneous detection, as the system struggles with unstable image feature values from varying angles, especially when images from various angles are not readily available for registration.

Innovation Solution

The system calculates similarity between feature values extracted from image data and outputs the person with the most similar angle information, improving detection accuracy by considering the angle information most similar to the recorded person, even when images from various angles are not previously registered.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If images from various angles are prepared and registered in advance, then detection accuracy is improved, but device complexity and data storage requirements increase

Engineering Contradiction:
Improvedetection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary action by calculating the angle of appearance for each detected person and selecting the most appropriate registered image based on angle matching before final detection. This preliminary angle-based selection prepares the comparison process in advance, improving detection accuracy without requiring all possible angle images to be stored

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system applies local quality by registering multiple images of the same person taken from different angles (each with its own angle information), and then locally selecting the most appropriate image based on the detected person's angle of appearance. This ensures that the comparison uses the most relevant image for that specific viewing angle

Inventive Principle:
Principle #3Local quality

2Measurement precision

If images from various angles are prepared and registered, then detection accuracy is improved, but data storage requirements increase

Engineering Contradiction:
Improvedetection accuracyVSAvoiddata storage
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The system changes the parameter approach by associating angle information with each registered person image, and then using this angle parameter to dynamically select the most appropriate image for comparison. This allows the system to handle various angles efficiently without storing exhaustive angle coverage, reducing data storage requirements while maintaining detection accuracy

Inventive Principle:
Principle #35Parameter changes

3Productivity

If the system processes images at high frame rates, then real-time detection capability is improved, but computational load and processing time increase

Engineering Contradiction:
Improvereal-time detection capabilityVSAvoidprocessing time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The system applies partial action by not processing all registered images for comparison, but instead selecting only the most appropriate image based on angle matching. This partial selection process reduces computational load significantly while still providing accurate real-time detection, avoiding the excessive processing that would result from comparing against all registered images

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3355269B1Specific person detection system and specific person detection method
Publication Date: 2023.08.02 HITACHI KOKUSAI ELECTRIC INC
  • EP3355269B1 patent drawingFigure 1~2
  • EP3355269B1 patent drawingFigure 3
  • EP3355269B1 patent drawingFigure 4A

AI summary

This specific person detection system: identifies, from among the persons recorded in a specific person recording unit, a person who most closely matches a feature value extracted from image data; calculates the degree to which feature values of a plurality of persons extracted from other image data match the identified person; and outputs, as an identification result, information about a person who has a feature value closely matching the identified person, and who is associated with angle information that most closely matches angle information associated with the identified person.