Facet-Based Image Search Using Convolutional Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual location and tracking of objects-of-interest in video surveillance systems is inefficient due to the large volume of concurrently recorded video, necessitating automated search functionalities for improved efficiency.
Innovation Solution
A computer-implemented method using a convolutional neural network (CNN) for automated image search, where the network is trained with state data indexed by facets such as age, gender, clothing, and vehicle characteristics, allowing for facet-based searches and weighting of search results by frequency, and enabling the system to revert to previous states for user-specific training and search optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual location and tracking of objects-of-interest is used in video surveillance systems, then system simplicity is maintained, but search efficiency and productivity are severely reduced due to large volume of concurrently recorded video
Solution Approach 1:
The patent replaces manual mechanical searching through video playback with an automated image recognition system using artificial neural networks. The system automatically processes video frames, identifies objects-of-interest based on learned features, and retrieves relevant segments, substituting human visual search with computational image analysis to dramatically improve productivity
Solution Approach 2:
The system enables self-service through automated training and search capabilities. Users can train the neural network by providing example images of objects-of-interest, and the system automatically learns to identify these objects without requiring manual configuration or intervention during operation, allowing the system to serve itself in improving search efficiency
2Measurement precision
If automated image search using artificial neural networks is implemented, then search accuracy and object identification precision are improved, but computational complexity and processing requirements increase
Solution Approach 1:
The system performs preliminary action by pre-training artificial neural networks with large datasets of example images before deployment. The training phase extracts and stores learned features and patterns in advance, so that during actual search operations, the pre-trained network can quickly and accurately identify objects without requiring complex real-time computations, thereby improving accuracy while managing computational complexity
Solution Approach 2:
The patent segments the video surveillance data into individual frames or segments for processing. The system processes video data in discrete units, allowing the neural network to analyze specific frames independently and efficiently. This segmentation enables parallel processing and reduces the computational burden compared to analyzing entire video streams continuously, improving both accuracy and efficiency
3Measurement precision
If facet-based training data organization is used, then search relevance and result quality are improved, but data processing complexity and storage requirements increase
Solution Approach 1:
The system applies parameter changes by organizing training data according to multiple facets or attributes (such as object type, location, time, characteristics). This multi-faceted organization allows the neural network to learn and search based on different parameters simultaneously, improving search relevance by enabling queries on multiple dimensions while efficiently managing data through structured parameter-based indexing
4Adaptability or versatility
If multiple user states are maintained for different users, then user-specific search optimization and adaptability are improved, but system complexity and processing overhead increase
Solution Approach 1:
The patent segments the system state into multiple user-specific states, where each user has their own trained neural network model or configuration parameters. This segmentation allows different users to have customized search capabilities optimized for their specific needs while maintaining manageable complexity through modular user state management, enabling adaptability without overwhelming system complexity
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, systems, and techniques for performing a facet search include receiving facet search commencement user input indicating that a search for a facet is to commence; in response to the facet search commencement user input, searching one or more video recordings for the facet; and displaying, on a display, facet image search results depicting the facet, wherein the facet image search results are selected from the one or more video recordings. An artificial neural network may be used for the facet search, and that network may be trained by generating a facet image training set that comprises training images, with the training images depicting a type of facet common to the training images; and training, by using the facet image training set, that neural network to classify the type of facet when a sample image comprising the type of facet is input to that network.