Object Recognition Region Extraction for Faster Small-Object Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition technologies using machine learning face challenges in maintaining accuracy and speed, particularly when dealing with high-resolution images, as reducing image size for input to machine learning models leads to information loss and increased inference time, especially for small objects.
Innovation Solution
An object inference module processes images to extract object-containing regions by calculating background and boundary lines, converting to binary images, and clustering these regions for input to a machine learning model, enhancing accuracy and reducing inference time by focusing on relevant image portions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the entire high-resolution image is input to the machine learning model, then the object recognition accuracy is improved, but the inference time increases
Solution Approach 1:
The image processing is divided into multiple stages: initial low-resolution analysis to identify object-containing regions, followed by extracting and processing only those specific regions at higher resolution. This segmentation approach allows the system to maintain high accuracy for objects while reducing overall processing time by excluding background areas.
Solution Approach 2:
The patent extracts only the object-containing regions from the full image for detailed processing. By calculating background lines and identifying regions above these lines (sky space), the system isolates and processes only relevant portions of the image, eliminating unnecessary background processing while preserving object detection accuracy.
2Productivity
If the image size is reduced for input to the model, then the inference time is reduced, but the object recognition accuracy deteriorates
Solution Approach 1:
The system processes images at different resolutions for different purposes: low-resolution for rapid initial analysis and region identification, high-resolution only for the extracted object-containing regions. This multi-resolution segmentation maintains inference speed while preserving accuracy for critical object areas.
Solution Approach 2:
Different parts of the image are processed with different quality levels. The background areas are processed at low resolution or excluded entirely, while object-containing regions are processed at high resolution. This local quality differentiation optimizes the balance between processing speed and recognition accuracy.
3Loss of time
If object-containing regions are extracted and processed separately, then the inference time is reduced, but the device complexity increases
Solution Approach 1:
The system performs preliminary processing steps before main object recognition: converting to grayscale, calculating background lines, and identifying object-containing regions. These preliminary actions simplify the subsequent main processing by pre-segregating relevant areas, reducing the overall computational complexity despite adding initial steps.
Data Source
AI summary
Provided is an apparatus for recognizing an object that includes an object inference module configured to process an original image captured by a camera module and generate an image of a size to be input to a machine learning inference model, wherein the object inference module includes the machine learning inference model, and outputs a result of recognition and classification of an object inferred through the machine learning inference model, and the machine learning inference model processes an input image to infer an object included in the input image.