Feature Interpolation for Object Detection Speed
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection methods are bottlenecked by the time-consuming process of extracting a high number of multi-scale features from images, such as 200,000 features in face detection, which hampers speed and efficiency.
Innovation Solution
Feature interpolation techniques are employed, where a portion of features are extracted and quantized into visual words, and then used to interpolate the remaining features using a spatial configuration database, significantly reducing the number of features needed while maintaining accurate object detection performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a high number of multi-scale features are extracted from images, then object detection accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent segments the feature extraction process into two parts: extracting a subset of features using conventional methods and interpolating the remaining features using a pre-computed spatial configuration database. This segmentation allows the system to achieve high detection accuracy without extracting all possible features, thus reducing processing time.
Solution Approach 2:
The patent performs preliminary action by pre-computing and storing spatial configuration databases during a training phase. These databases contain information about the spatial relationships between visual words in training images. During the detection phase, this pre-computed information is used to interpolate features, eliminating the need to extract all features from scratch and significantly reducing processing time.
2Productivity
If the number of extracted features is reduced, then processing speed is improved, but detection performance may deteriorate
Solution Approach 1:
The patent introduces an intermediary mechanism in the form of a spatial configuration database that mediates between the extracted features and the complete feature set. This database contains pre-computed spatial relationships from training images and is used to interpolate missing features during detection, ensuring that reduced feature extraction does not compromise detection performance.
Solution Approach 2:
The patent uses copying by creating a synthetic representation of the complete feature set through interpolation. Instead of extracting all features directly, the system copies the spatial configuration patterns from training images to generate the remaining features, maintaining detection accuracy while reducing computational burden.
Data Source
AI summary
Feature interpolation techniques are described. In a training stage, features are extracted from a collection of training images and quantized into visual words. Spatial configurations of the visual words in the training images are determined and stored in a spatial configuration database. In an object detection stage, a portion of features of an image are extracted from the image and quantized into visual words. Then, a remaining portion of the features of the image are interpolated using the visual words and the spatial configurations of visual words stored in the spatial configuration database.