Real-Time Object Recognition via Dimensional Feature Vector Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques fail to recognize a search object in a video in real time, despite advancements in image recognition and feature clustering.
Innovation Solution
A system that includes a first local feature storing unit, a second local feature generating unit, and a recognizing unit, which stores and generates feature vectors for local areas in images, and recognizes the search object by selecting a smaller number of dimensions to match a prescribed ratio of features between the search object and video frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image recognition techniques are used to search for objects in videos, then recognition accuracy can be maintained, but real-time recognition cannot be achieved
Solution Approach 1:
The patent segments the image recognition process by dividing features into local features (extracted from specific regions) and global features (overall image characteristics). This segmentation allows the system to process only relevant local regions rather than entire images, significantly reducing computation time while maintaining recognition accuracy through focused feature analysis.
Solution Approach 2:
The patent extracts only the essential local features from video frames that are most relevant for object recognition, rather than processing all image data. By taking out and focusing on discriminative local features, the system achieves real-time processing speed while preserving the key information needed for accurate recognition.
2Measurement precision
If comprehensive feature vectors with many dimensions are used for object recognition, then recognition accuracy improves, but processing time increases
Solution Approach 1:
The patent extracts and retains only the most informative dimensions from comprehensive feature vectors. By identifying and extracting the key dimensions that contribute most to recognition accuracy, the system reduces the feature vector size and processing time while maintaining the essential discriminative power needed for accurate object recognition.
Solution Approach 2:
The patent dynamically adjusts the number of dimensions retained in feature vectors based on recognition requirements. By changing the parameter of feature vector dimensionality, the system can optimize between accuracy and processing speed, keeping high-dimensional features when accuracy is critical and reducing dimensions when real-time processing is needed.
Data Source
AI summary
A search object and m-number of first local features respectively constituted by a feature vector of 1 to i dimensions of local areas of m-number of feature points in an image of the search object are stored, feature points are extracted from the image, second local features respectively constituted by a feature vector of 1 dimension to j dimensions are generated with respect to local areas of n-number of feature points, a smaller number of dimensions among the number of dimensions i of the first local features and the number of dimensions j of the second local features is selected, and an existence of the search object in the image in the video is recognized when a prescribed ratio of the m-number of first local features up to the selected number of dimensions corresponds to the n-number of second local features up to the selected number of dimensions.