Train Interior Object Detection Using Unsupervised Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting objects within the interior of public transport vehicles that are not part of the furnishings, such as people or items, suffer from inaccuracies and require manual labeling of training images, making them inefficient and unreliable, especially under varying lighting conditions.
Innovation Solution
A computer-implemented method using a convolutional autoencoder trained through unsupervised learning to reconstruct interior images without objects, comparing pixel differences to detect deviations, and employing a classifier to enhance detection accuracy, with optional preprocessing to reduce image size and edge regions, and utilizing a second neural network for supervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network is trained using supervised learning with manually labeled training images, then the detection accuracy can be improved, but the complexity and time required for training increases significantly
Solution Approach 1:
The patent applies self-service by using unsupervised learning where the neural network automatically learns to distinguish between interior images with and without objects without requiring manual labeling. The system trains on raw images and autonomously identifies patterns, eliminating the need for human annotation while maintaining detection accuracy.
Solution Approach 2:
The patent replaces the mechanical process of manual image labeling with an automated unsupervised learning mechanism. Instead of requiring human experts to label training images, the system uses algorithmic learning to automatically extract features and patterns from raw images, substituting manual labor with computational intelligence.
2Reliability
If manual labeling of training images is performed, then the training data quality improves, but the time and resources required for preparation increases
Solution Approach 1:
The system performs self-service by automatically generating and utilizing training data without human intervention. The unsupervised learning algorithm autonomously processes raw images, identifies relevant patterns, and creates training datasets, eliminating the time-consuming manual labeling process while maintaining data quality through algorithmic consistency.
Solution Approach 2:
The patent applies preliminary action by pre-processing raw images automatically before training. The system performs initial image analysis, feature extraction, and data preparation in advance through automated algorithms, eliminating the need for manual data preparation and reducing the overall training preparation time.
3Measurement precision
If the neural network processes all pixels in the image, then the detection completeness improves, but the processing time and computational resources increase
Solution Approach 1:
The patent applies segmentation by dividing the image processing into functional segments: the neural network processes only relevant pixel regions that contain object information, while excluding redundant background areas. This selective processing maintains detection completeness for objects of interest while significantly reducing overall computational load and processing time.
Solution Approach 2:
The system implements local quality by applying different processing intensities to different regions of the image. Areas containing potential objects receive detailed analysis, while uniform background regions are processed more lightly or skipped entirely, optimizing the balance between detection completeness and processing speed.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A computer-implemented method for detecting an object in the interior of a public transport vehicle that is not part of the interior furnishings of the public transport vehicle, wherein a camera captures an image of the interior of the public transport vehicle, wherein the captured image is converted into a reconstructed image using a trained neural network, wherein the neural network was trained with images of the interior of the public transport vehicle using a self-learning method, wherein the reconstructed image is compared with the captured image, wherein a pixel-by-pixel comparison is performed when comparing the captured image with the reconstructed image, and wherein a brightness difference and/or a color difference between the pixels is determined when comparing the pixels to be compared in the captured image and the reconstructed image.wherein a pixel of the reconstructed image is recognized as a mismatched pixel if the brightness difference and/or the color difference exceeds a threshold, wherein, based on the mismatched pixels, an object is detected in the interior of the public transport vehicle that is not part of the interior furnishings of the public transport vehicle, and wherein a signal is output upon detection of the object that is not part of the interior furnishings of the public transport vehicle.