Image Processing System with Self-Labeling for Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image processing systems face challenges in robustness, workability, and detection accuracy when detecting objects from images, particularly due to changes in brightness and the need for extensive manual labeling in deep learning setups.
Innovation Solution
An image processing system that combines a first detector using a model pattern to detect objects and a learning device that learns a model from the first detector's results, allowing for improved robustness and accuracy without requiring extensive manual labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning is used to improve detection robustness, then detection accuracy is improved, but extensive manual labeling is required which reduces workability
Solution Approach 1:
The system performs automatic self-labeling by using the first detector's detection results to automatically generate training data labels. The learning device automatically creates labeled training images from the captured images and detection results, eliminating the need for manual labeling by users. This self-service mechanism resolves the contradiction by maintaining deep learning's robustness while removing the manual labeling burden.
2Measurement precision
If manual labeling is performed to generate training data, then detection accuracy is improved, but the setting work becomes more complex and time-consuming
Solution Approach 1:
The system performs preliminary detection using the first detector with model patterns to identify object positions and features before generating training data. By pre-detecting objects and their characteristics, the system prepares labeled training data automatically before the learning process, eliminating the need for time-consuming manual labeling while ensuring accurate detection results are captured for training.
3Ease of operation
If model pattern detection is used to improve workability, then ease of operation is improved, but detection accuracy decreases when features become invisible
Solution Approach 1:
The system merges two detection approaches: model pattern-based detection (first detector) and learning model-based detection (second detector). The model pattern approach provides ease of operation and quick setup, while the learning model approach provides robustness when features are invisible or degraded. By combining both methods, the system achieves both ease of operation and reliable detection under various conditions.
Data Source
AI summary
An image processing system that detects a picture of an object from an image in which the object is captured includes: a first detector that detects a picture of the object from the image based on a model pattern representing a feature of the picture of the object; a learning device that learns a learning model using the image used for detection by the first detector as input data, and using a detection result by the first detector as training data; and a second detector that detects the picture of the object from the image based on the learning model learned by the learning device.