Intruder Detection Using Viewpoint-Specific CNNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Video surveillance systems face challenges in accurately detecting intruders and estimating abnormal behavior due to varying viewpoints from multiple cameras, which can lead to increased error rates and reduced security effectiveness.
Innovation Solution
The system employs a method involving multiple convolutional neural networks, each specialized for different viewpoints, to extract feature maps and combine them using weighted sums, allowing for accurate intruder detection and behavior estimation by classifying and transforming images based on object skeletons and body part ratios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If multiple cameras are installed at different heights and positions to monitor a wide area, then the monitored area coverage is improved, but the viewpoint differences cause increased detection error rates
Solution Approach 1:
The patent divides the image processing task into multiple segments by creating separate convolutional neural networks for different viewpoint categories (front view, lateral view, planar view, rear view). Each CNN is specialized to process images from its specific viewpoint, allowing accurate feature extraction from heterogeneous camera angles while maintaining high detection precision across the entire monitored area.
Solution Approach 2:
The patent applies local quality by assigning different processing characteristics to different parts of the input data. Specifically, images are classified into viewpoint categories and processed by corresponding specialized CNNs, with each network having optimized features for its specific viewpoint type. This allows the system to maintain high detection accuracy for each local viewpoint while covering the overall monitored area.
2Length of moving object
If cameras are positioned at high locations to detect intruders from long distance, then the detection range is improved, but the varying installation heights create more diverse image viewpoints and postures
Solution Approach 1:
The patent segments the complex processing task by dividing it into multiple specialized convolutional neural networks, each handling a specific viewpoint category. This segmentation allows the system to manage long-distance detection from various heights by processing each viewpoint type with optimized features, reducing the overall complexity compared to using a single general-purpose network.
Solution Approach 2:
The patent changes the parameter of viewpoint perspective by classifying images into different viewpoint categories (front view, lateral view, planar view, rear view) and processing them with corresponding specialized networks. This parameter-based organization allows the system to handle diverse installation heights and camera positions effectively, managing processing complexity through structured parameter differentiation.
3Device complexity
If heterogeneous images with different viewpoints are processed by a single neural network, then the system simplicity is maintained, but the detection error rate increases
Solution Approach 1:
The patent segments the neural network into multiple specialized convolutional networks, each dedicated to processing a specific viewpoint category. This segmentation improves detection reliability by ensuring that each network develops optimized features for its specific viewpoint type, rather than a single network attempting to handle all viewpoints with reduced effectiveness for each.
Solution Approach 2:
The patent implements local quality by giving each viewpoint category its own specialized processing network with optimized features. Front view images are processed by front-view-specialized networks, lateral views by lateral-view networks, and so on. This ensures high detection reliability for each local viewpoint type while maintaining an organized system structure through clear functional specialization.
Data Source
AI summary
An exemplary embodiment provides an intruder detection method capable of accurately detecting an intruder and estimating an abnormal behavior of the intruder even when viewpoints of acquired images are different from each other. An intruder detection method is suitable for being performed by an intruder detection device for detecting an intruder based on images and includes: receiving input images acquired by multiple cameras; extracting feature maps associated with a plurality of viewpoints by applying the input images to a plurality of convolutional neural networks provided separately for the plurality of viewpoints of the images; and detecting the intruder based on the feature maps associated with the plurality of viewpoints.