Object Recognition via Vocabulary Tree Density Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition methods fail to accurately identify objects in input images, especially when parts are occluded or when recognizing multiple objects, and are not suitable for mobile applications due to large data sizes required for training.
Innovation Solution
A method that involves obtaining feature points and descriptors from input images, determining indices using a pretrained vocabulary tree, estimating density distributions, and recognizing objects based on these indices, which includes calculating similarity scores and using kernel density estimation for segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional object recognition methods are used, then object recognition accuracy may be maintained in simple cases, but recognition fails when objects are occluded or multiple objects need to be recognized
Solution Approach 1:
The patent segments objects into multiple parts (occluded parts and visible parts) and processes them separately through different neural network branches. The first neural network processes occluded parts while the second neural network processes visible parts, allowing the system to handle occlusion and multiple objects effectively without compromising recognition accuracy.
2Measurement precision
If numerous training images are used to accurately recognize objects, then recognition accuracy improves, but data size becomes too large for mobile applications
Solution Approach 1:
The patent segments the training process by using two separate neural networks that are trained on different types of image parts (occluded and visible). This segmentation allows the system to achieve accurate recognition without requiring a single large-scale training dataset, making it more suitable for mobile applications with limited storage capacity.
Solution Approach 2:
The patent creates a universal object recognition system that can handle multiple scenarios (single objects, multiple objects, occluded objects, visible objects) using a unified dual-branch architecture. This multi-functional approach eliminates the need for separate specialized models for different object conditions, reducing overall data requirements while maintaining versatility.
Data Source
AI summary
A method and apparatus for recognizing an object may obtain, from an input image, feature points and descriptors corresponding to the feature points, determine indices of the feature points based on the descriptors, estimate a density distribution of feature points for each of the indices, and recognize an object included in the input image based on the estimated density distribution.