Annotation Device for Object Detection via Multi-Angle Image Capture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting objects from images face challenges due to variability in brightness and positional relationships between the camera and object, leading to inaccurate object detection, especially when relying on user annotation for deep learning models.
Innovation Solution
An annotation device that includes an image-capturing device, a robot for adjusting the object's position, and a control unit to capture images from various angles, allowing for accurate annotation processing by converting object positions between different coordinate systems and storing these positions with corresponding images, facilitating robust object detection through supervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If user performs annotation processing manually, then annotation can be done with some accuracy, but the accuracy is reduced due to variations in user references and the process is time-consuming
Solution Approach 1:
The system enables self-service annotation by allowing the object to annotate itself. The annotation information generation unit automatically generates annotation data based on images captured from multiple angles and positions, eliminating the need for manual user annotation while maintaining high accuracy through multi-perspective validation.
Solution Approach 2:
The patent replaces the mechanical manual annotation process with an automated image processing system. The annotation information generation unit uses computer vision and image processing algorithms to automatically extract and annotate object features, substituting human manual work with automated computational methods.
2Reliability
If machine learning is used for object detection, then detection can be performed more robustly, but a large number of annotated images are required for supervised learning
Solution Approach 1:
The system segments the annotation process by capturing images from multiple angles and positions separately. This segmentation allows the generation of diverse training data in an organized manner, providing the quantity of training images needed for robust machine learning while maintaining systematic control over the data collection process.
Solution Approach 2:
The patent adds dimensional diversity to training data by capturing images from multiple angles (horizontal, vertical, inclined) and positions. This multi-dimensional approach enriches the training dataset without requiring a proportional increase in the number of different objects, thereby improving detection robustness efficiently.
3Device complexity
If images are captured from fixed positions, then the capture process is simple, but the positional relationship between camera and object changes cause detection failures
Solution Approach 1:
The system implements dynamic image capture by automatically varying the camera's position and angle. The image capturing unit captures images from multiple predetermined positions and angles, creating a dynamic data collection process that adapts to different object orientations and ensures reliable detection regardless of positional relationships.
Solution Approach 2:
The annotation system achieves universality by capturing images from multiple angles and positions, making the annotation process applicable to objects in various orientations and configurations. This multi-functional approach ensures that the system can handle diverse detection scenarios without requiring separate processes for each case.
Data Source
AI summary
An annotation device includes an image-capturing device, a robot, a control unit, a designation unit, a coordinate processing unit, and a storage unit. The control unit controls the robot so as to acquire a learning image of a plurality of objects, each having a different positional relationship with the image-capturing devices. Furthermore, the storage unit converts a position of the object in a robot coordinate system into a position of the object in an image coordinate system at the time of image-capturing or a position of the object in a sensor coordinate system, and stores the position thus converted together with the learning image.