Method of training object recognition model by using spatial information and computing device for performing the method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural network models for object recognition, such as those used in robot vacuum cleaners, face accuracy issues due to significant illumination differences between daytime spatial modeling and nighttime operational environments, leading to misrecognition of objects.
Innovation Solution
A method is developed to train neural network models using spatial information, including illumination data from multiple spots in a space, by generating training data through style transfer of images captured during different lighting conditions, which are then used to improve object recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If spatial modeling is performed during daytime with lights on, then the map generation is accurate, but object recognition fails during nighttime operations when lights are off due to illumination differences
Solution Approach 1:
The patent captures reference images during daytime spatial modeling when illumination is stable and objects are clearly visible. These reference images are stored and later used for comparison during nighttime operations, allowing the system to maintain accurate object recognition despite changes in illumination conditions.
Solution Approach 2:
The patent creates a visual copy of the spatial environment during daytime by capturing and storing reference images. These copied images serve as a template for object recognition during nighttime, enabling the system to identify objects by comparing current low-light images against the stored daytime reference copies.
2Productivity
If the neural network model is trained only on daytime images, then training is simple and fast, but recognition accuracy drops significantly in low-light nighttime conditions
Solution Approach 1:
The patent performs preliminary capture of reference images during daytime spatial modeling before nighttime operations begin. This preliminary action ensures that training data reflecting actual nighttime illumination conditions is available, allowing the neural network to be trained on both daytime and nighttime images for improved reliability without significantly increasing training complexity.
3Measurement precision
If multiple images under different illumination conditions are used for training, then recognition accuracy under varying lighting improves, but data collection and processing complexity increases
Solution Approach 1:
The patent merges the daytime spatial modeling process with reference image capture. By combining these functions, the system collects training data during normal operations without requiring separate data collection missions, thus improving recognition accuracy while minimizing additional complexity in the data collection process.
Data Source
AI summary
A method of training an object recognition model by using spatial information is provided. The method includes obtaining spatial information including illumination information corresponding to a plurality of spots in a space, obtaining illumination information corresponding to at least one spot of the plurality of spots from the spatial information, obtaining training data by using the obtained illumination information and an image obtained by capturing the at least one spot, and training a neural network model for object recognition by using the training data.