Image Object Detection and Shape Modeling with Joint Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting and modeling objects and characteristic regions in images suffer from inaccuracies due to the use of characteristic points, leading to incomplete or inconsistent annotations, especially in cases where objects are partially concealed.
Innovation Solution
A machine learning system is trained using augmented reality images that include virtual elements representing the objects and characteristic regions, employing a joint learning process to accurately segment and parametrize these elements, using a cost function that combines cross-entropy and Euclidean distance to enhance detection and modeling accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If characteristic points are used for detecting objects and characteristic regions, then the detection process is simplified, but the accuracy of detection and modeling deteriorates
Solution Approach 1:
The patent segments the detection task into two parts: first detecting characteristic points using simple methods, then generating a segmentation model that provides precise boundary information. This segmentation approach allows the system to maintain operational simplicity while achieving high detection accuracy through the additional segmentation layer.
Solution Approach 2:
The patent introduces a segmentation model as an intermediary between characteristic point detection and final object modeling. This intermediary component transforms the imprecise characteristic points into accurate boundary representations, resolving the contradiction between simple detection processes and accurate results.
2Measurement precision
If manual annotation is used to generate characteristic points, then some accuracy is improved, but inaccuracy and inconsistency remain especially when objects are concealed
Solution Approach 1:
The system employs self-service through automated segmentation model generation that does not require manual annotation. The segmentation model automatically adapts to various scenarios including concealed objects, eliminating human error and inconsistency while maintaining high accuracy. This self-service approach ensures reliable and consistent results across different detection scenarios.
3Productivity
If characteristic points are used for modeling, then the process is faster, but the shape representation becomes incomplete or inconsistent
Solution Approach 1:
The patent merges characteristic point detection with segmentation model generation into a unified process. By combining these two approaches, the system achieves both the speed of characteristic point methods and the accuracy of segmentation-based shape representation, resolving the contradiction between productivity and shape quality.
Solution Approach 2:
The system performs preliminary action by pre-processing images to generate segmentation models before final detection and modeling. This preliminary segmentation step ensures that accurate shape information is available upfront, allowing fast subsequent processing while maintaining high shape representation quality.
Data Source
AI summary
A learning method of a machine learning system carries out the steps of: obtaining augmented reality images including a real image and at least one virtual element representative of the object and/or the characteristic region; obtaining, for each augmented reality image, learning information including, for at least one given virtual element of the augmented reality image: a model for segmenting the given virtual element, and a set of contour points corresponding to a parameterisation of the given virtual element; and learning on the basis of the plurality of augmented reality images and the learning information, delivering a set of parameters enabling the machine learning system to detect and model the object and/or the characteristic region in a given image.