Segmented Object Detection for UHD Images With Lower Compute Load

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object detection methods struggle with high calculation loads and missed detections in ultra-high-definition images, particularly in deep learning-based systems like YOLOv3, due to the division of images into numerous segments, leading to unstable detection results and increased computational demands.

Innovation Solution

An object detection device that divides input images into segments, performs object detection on reduced whole images and segmented images, and synthesizes results, while dynamically controlling the frame rate for each segment based on estimated object fluctuations to reduce calculation load and maintain stability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the input image is divided into multiple segmented images for object detection, then small objects can be detected more accurately, but the calculation amount increases enormously

Engineering Contradiction:
Improvedetection accuracyVSAvoidcalculation amount
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies segmentation by dividing the input image into multiple segmented images (e.g., 9 segments for 4K resolution) to enable detection of both large and small objects. Each segmented image is processed separately through the object detection model, allowing small objects to be detected with sufficient detail while managing computational load through selective processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements periodic action by performing object detection on segmented images at specific time intervals (every N frames) rather than continuously for every frame. This reduces the calculation amount by processing only a subset of frames while still maintaining detection capability through temporal sampling.

Inventive Principle:
Principle #19Periodic action

2Stability of the object's composition

If object detection is performed on every frame of segmented images, then detection stability is improved, but the calculation load becomes enormous

Engineering Contradiction:
Improvedetection stabilityVSAvoidcalculation amount
Core Design Contradiction:
Stability of the object's compositionVSQuantity of substance

Solution Approach 1:

The patent performs object detection on segmented images periodically every N frames rather than on every frame. This reduces the calculation load by a factor of N while maintaining detection stability through temporal sampling and result synthesis across multiple frames.

Inventive Principle:
Principle #19Periodic action

Solution Approach 2:

The patent merges detection results from multiple segmented images and multiple time points to achieve stable object detection. By synthesizing results from different segments and frames, the system maintains detection stability without requiring processing of every single frame.

Inventive Principle:
Principle #5Merging (Combining)

3Speed

If the input image is reduced to a restricted size for object detection, then processing speed increases, but small objects become difficult to detect

Engineering Contradiction:
Improveprocessing speedVSAvoiddetection accuracy
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The patent segments the reduced-size input image into multiple smaller regions, allowing the detection model to process each segment at full resolution. This enables small objects to be detected with sufficient detail while the overall processing speed is maintained through the reduced input size.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by processing different regions of the image at different effective resolutions. Small objects in specific segments are detected with higher effective resolution than would be possible in a uniformly reduced image, while maintaining overall processing efficiency.

Inventive Principle:
Principle #3Local quality

4Adaptability or versatility

If both overall processing and division processing are performed to detect large and small objects, then detection coverage is improved, but the system complexity increases

Engineering Contradiction:
Improvedetection coverageVSAvoidsystem complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent makes the segmented image processing system universal by designing it to handle both large and small objects through the same object detection model. The same model processes both full-frame and segmented images, eliminating the need for separate specialized processors for different object sizes.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent merges the results from overall processing and division processing into a unified detection output. By synthesizing results from both approaches, the system achieves comprehensive detection coverage for all object sizes while presenting a single integrated result to the user.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP4354385B1Object detection device and method
Publication Date: 2026.02.11 NT T INC
  • EP4354385B1 patent drawingFigure 1
  • EP4354385B1 patent drawingFigure 2
  • EP4354385B1 patent drawingFigure 3

AI summary

An object detecting device includes: an overall processing unit (3) configured to perform object detection processing on an input image; an allocation processing unit (5) configured to determine a segmented image for each frame which is subject to object detection among the plurality of segmented images in advance in a cycle of N frames, and reserve the determined segmented image; a selection processing unit (2) configured to select and output, for each frame, some of the segmented images that are subject to object detection among the plurality of segmented images based on a reservation result from the allocation processing unit (5); a division processing unit (4) configured to perform object detection processing on each of the segmented images output from the selection processing unit (2); and a synthesizing processing unit (8) configured to synthesize an object detection result from the overall processing unit (3) and an object detection result from the division processing unit (4).