Person Detection Frame Sizing for Accurate Image Position Display

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing technologies fail to accurately represent the size of a frame line superimposed on a person in a captured image, leading to misinterpretation of the person's position due to the frame line's size being fixed or proportional to the person's size, which can incorrectly suggest the person's distance based on their height.

Innovation Solution

A picture recognition apparatus that adjusts the size of the frame line based on the detected person's height and distance, enlarging the frame line for shorter individuals to prevent misinterpretation of their position relative to the camera.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep learning models with many parameters are used to improve recognition accuracy, then recognition precision improves, but the number of parameters increases leading to larger model size and higher computational requirements

Engineering Contradiction:
Improverecognition precisionVSAvoidnumber of parameters
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts and removes redundant parameters from the deep learning model through regularization techniques. L1 regularization identifies and eliminates unnecessary weight parameters, while dropout selectively deactivates redundant neurons during training. This extraction process reduces the total parameter count while preserving the essential features needed for accurate recognition.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transforms the model architecture by changing parameter representations and distributions. Weight decay modifies parameter values toward smaller magnitudes, sparse optimization encourages sparse parameter distributions, and quantization changes parameter precision from high-precision floats to lower-precision representations. These parameter changes reduce model size and complexity while maintaining recognition precision.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If high-resolution images are used to improve recognition accuracy, then recognition precision improves, but processing time increases

Engineering Contradiction:
Improverecognition precisionVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments high-resolution images into multiple lower-resolution patches or regions before processing. Instead of processing the entire high-resolution image at once, the model divides it into manageable segments, processes each segment independently or with reduced computational resources, and then aggregates the results. This segmentation maintains recognition precision by preserving local features while significantly reducing processing time and computational load.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If more training data is collected to improve model accuracy, then recognition precision improves, but data collection time and storage requirements increase

Engineering Contradiction:
Improvemodel accuracyVSAvoiddata collection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent creates synthetic copies of training data through data augmentation techniques. Instead of collecting more real-world data, the system generates synthetic variations of existing training samples by applying transformations such as rotations, flips, color adjustments, and noise additions. These synthetic copies expand the effective training dataset size, improving model accuracy and generalization without requiring additional data collection time or storage resources.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4227895B1Image recognition device, image recognition method, and program
Publication Date: 2026.04.29 JVC KENWOOD CORP
  • EP4227895B1 patent drawingFigure 1
  • EP4227895B1 patent drawingFigure 2~3
  • EP4227895B1 patent drawingFigure 4

AI summary

A picture recognition apparatus 10 includes: a person detector 14 that detects a person included in a captured picture and specifies a detection position and a detection size of the person in the captured picture; a frame line determiner 16 that determines a size of a frame line different from the detection size on the basis of the specified detection position of the person; a picture generator 18 that generates a display picture in which the frame line of the determined size is superimposed on the detection position of the captured picture; and a display controller 20 that causes a display device 28 to display the generated display picture.