Object Detection-Guided Image Segmentation for Real-Time Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies for automated driving and self-propelled robots face high processing loads when estimating distances to objects, particularly with techniques like semantic segmentation, which are not suitable for real-time applications.
Innovation Solution
An image processing device that combines object detection with semantic segmentation by adjusting detection regions based on expected correct-answer regions, using neural networks to enhance accuracy while minimizing processing load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If semantic segmentation is performed on the entire image to achieve high-accuracy region division, then measurement precision is improved, but processing time increases and productivity deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the image processing into two stages: first performing object detection to identify candidate regions, then performing semantic segmentation only on those detected regions. This segmented approach maintains high accuracy where needed while reducing overall processing load compared to full-image semantic segmentation.
Solution Approach 2:
The patent implements local quality by applying different processing methods to different parts of the image: object detection is applied to the entire image to locate potential objects, while semantic segmentation is applied locally only to the detected object regions. This ensures high accuracy for region division while minimizing processing time by avoiding unnecessary computation in background areas.
2Measurement precision
If both object detection and semantic segmentation models are implemented to achieve high-accuracy region division, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent merges object detection and semantic segmentation into a unified processing pipeline where the output of object detection directly feeds into semantic segmentation for detected regions. This integration reduces device complexity by coordinating the two models systematically rather than maintaining them as separate, independent systems.
Solution Approach 2:
The patent applies preliminary action by performing object detection first to identify and locate objects before applying semantic segmentation. This preliminary step reduces the input scope for the semantic segmentation model, simplifying its configuration and reducing computational requirements while maintaining high accuracy.
3Measurement precision
If semantic segmentation is performed on the entire image to achieve high-accuracy region division, then measurement precision is improved, but processing load increases and energy consumption increases
Solution Approach 1:
The patent segments the processing workload by first identifying object regions through detection, then applying the energy-intensive semantic segmentation only to those specific regions. This segmentation strategy maintains high measurement precision while dramatically reducing overall energy consumption compared to processing the entire image.
Solution Approach 2:
The patent applies local quality by concentrating computational resources and energy consumption on the detected object regions where high accuracy is needed, rather than uniformly processing the entire image. This approach achieves high measurement precision with minimal energy expenditure.
Data Source
AI summary
In order to realize an image processing device capable of obtaining a high-accuracy region division result while suppressing a processing load, an image processing device includes an object detection unit configured to detect an object from image data, a region setting unit configured to set a target region in which a category is assigned to pixels in accordance with a detection result of the object detection unit, and a category assignment unit configured to assign the category to the pixels belonging to the category of the target region which is set by the region setting unit.