Video Object Recognition Using Hierarchical Dimensional Vector Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for recognizing objects in videos do not enable real-time notification of recognition results while maintaining recognition accuracy.
Innovation Solution
A video processing system with a first local characteristic quantity storing unit, a second local characteristic quantity generating unit, a recognizing unit, and a displaying unit, which stores and generates characteristic vectors for local areas in images, selects a smaller number of dimensions for recognition, and displays recognition information in real-time when a prescribed proportion of corresponding vectors match.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If clustering characteristic amounts is used to improve recognition speed, then recognition speed is improved, but real-time notification of recognition results cannot be achieved
Solution Approach 1:
The patent divides the characteristic vector into multiple dimensions and processes them hierarchically. The recognizing unit first compares lower-dimensional characteristic quantities, then progressively compares higher-dimensional quantities only when needed. This segmentation of the comparison process into stages enables faster recognition by avoiding unnecessary full-dimensional comparisons, thereby achieving real-time notification capability while maintaining recognition accuracy.
2Measurement precision
If full-dimensional characteristic vectors are compared for accurate recognition, then recognition accuracy is maintained, but processing time increases
Solution Approach 1:
The patent implements a dynamic comparison strategy where the recognizing unit adaptively determines the comparison depth based on the matching results at each dimension level. When lower-dimensional characteristic quantities show sufficient similarity, the unit terminates comparison early without processing higher dimensions. This dynamic adjustment of processing depth maintains recognition accuracy for clear matches while reducing processing time for ambiguous cases, resolving the contradiction between accuracy and speed.
Data Source
AI summary
The present invention is to notify a recognition result with respect to a recognition object in a video in real time while maintaining recognition accuracy. A recognition object and m-number of first local characteristic quantities which are respectively 1-dimensional to i-dimensional characteristic vectors are stored in association with each other, and n-number of second local characteristic quantities which are respectively 1-dimensional to j-dimensional characteristic vectors are generated for n-number of local areas respectively including n-number of characteristic points from an image in a video. In addition, a smaller dimension is selected from the dimension i and the dimension j, a recognition that the recognition object exists in an image in the video is made when it is determined that a prescribed proportion or more of the m-number of first local characteristic quantities which are characteristic vectors up to the selected number of dimensions correspond to the n-number of second local characteristic quantities which are characteristic vectors up to the selected number of dimensions, and information representing the recognition object is displayed in superposition on an image in which the recognition object exists in the video.