Video Object Identification Using Layered Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for object identification in content, such as images and video sequences, face challenges in terms of speed and accuracy.
Innovation Solution
A system utilizing a first neural network to detect the presence of a predetermined object and an ensemble of second neural networks to identify the specific object, employing convolutional neural networks (CNNs) trained on distinct datasets to enhance detection and classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single neural network is used for both detection and identification, then device complexity is reduced, but measurement precision and reliability deteriorate
Solution Approach 1:
The patent divides the object recognition task into two separate neural networks: a first neural network for detection (determining presence/absence) and a second neural network for identification (determining specific object type). This segmentation allows each network to be optimized for its specific function, improving overall accuracy while maintaining manageable complexity.
Solution Approach 2:
The patent introduces an intermediary component that receives output from the first neural network and conditions the input to the second neural network. This intermediary layer enables the system to efficiently route processing based on detection results, improving both accuracy and computational efficiency.
2Measurement precision
If multiple neural networks are used for detection and identification, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent divides the object recognition task into two separate neural networks: a first neural network for detection (determining presence/absence) and a second neural network for identification (determining specific object type). This segmentation allows each network to be optimized for its specific function, improving overall accuracy while maintaining manageable complexity.
Solution Approach 2:
The first neural network performs preliminary detection to determine whether an object is present before the second neural network performs identification. This preliminary action filters out cases where no object is present, preventing unnecessary processing and reducing overall system complexity.
3Measurement precision
If comprehensive object analysis is performed, then measurement precision improves, but productivity decreases
Solution Approach 1:
The first neural network performs preliminary detection to determine whether an object is present before the second neural network performs identification. This preliminary action filters out cases where no object is present, preventing unnecessary processing and improving processing speed while maintaining accuracy for positive cases.
Solution Approach 2:
The patent divides the object recognition task into two separate neural networks: a first neural network for detection (determining presence/absence) and a second neural network for identification (determining specific object type). This segmentation allows each network to be optimized for its specific function, improving overall accuracy while maintaining manageable complexity.
4Device complexity
If a simple detection system is used, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent divides the object recognition task into two separate neural networks: a first neural network for detection (determining presence/absence) and a second neural network for identification (determining specific object type). This segmentation allows each network to be optimized for its specific function, improving overall accuracy while maintaining manageable complexity.
Data Source
Figure 1
Figure 2
Figure 3~7
AI summary
A method for identifying an object within a video sequence, wherein the video sequence comprises a sequence of images, wherein the method comprises, for each of one or more images of the sequence of images: using a first neural network to determine whether or not an object of a predetermined type is depicted within the image; and in response to the first neural network determining that an object of the predetermined type is depicted within the image, using an ensemble of second neural networks to identify the object determined as being depicted within the image.