Variable-Scale Multi-Feature Fusion Convolutional Network for Pedestrian Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pedestrian segmentation methods based on single network structures struggle to accurately segment pedestrians of varying scales in intelligent roadside images, leading to boundary fuzziness and missed detections due to limited feature expression and adaptability.
Innovation Solution
A variable-scale multi-feature fusion convolutional network is designed, comprising two parallel neural networks that extract local and global features for different scales, with a two-level fusion strategy to combine features, enhancing the robustness and accuracy of pedestrian segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single network structure is used for pedestrian segmentation, then the device complexity is reduced, but the segmentation precision deteriorates due to inability to handle variable-scale pedestrians effectively
Solution Approach 1:
The patent divides the pedestrian segmentation task into two parallel network structures: a first network for extracting local features and a second network for extracting global features. This segmentation of the single network into multiple specialized networks allows each network to focus on specific feature types, improving segmentation precision for variable-scale pedestrians while managing complexity through functional division
Solution Approach 2:
The patent introduces a multi-scale feature extraction dimension by designing networks with different receptive fields and feature extraction capabilities. The first network focuses on local detailed features while the second network captures global contextual features, adding a dimensional aspect of feature scale that improves precision without simply increasing network depth
2Loss of information
If deep convolutional layers are increased to extract more features, then the feature expression ability is improved, but the boundary detection capability deteriorates due to loss of local detailed information
Solution Approach 1:
The patent segments feature extraction into two parallel pathways: one dedicated to local feature extraction and another to global feature extraction. This segmentation ensures that local boundary information is preserved in the first network while the second network provides global context, preventing the loss of boundary detection precision that occurs when using only deep convolutional layers
Solution Approach 2:
The patent merges the output features from the first network (local features) and the second network (global features) through feature fusion. This combination allows the system to retain both detailed local boundary information and comprehensive global context, achieving both strong feature expression and accurate boundary detection
3Adaptability or versatility
If the receptive field is expanded to capture global features, then the adaptability to large-scale pedestrians is improved, but the local detailed feature extraction deteriorates
Solution Approach 1:
The patent segments the feature extraction function into two specialized networks: the first network maintains a focused receptive field for precise local feature extraction, while the second network expands the receptive field to capture global features. This segmentation allows each network to excel at its specific function without compromising the other
Solution Approach 2:
The patent applies local quality by designing the first network with architecture optimized for local detailed feature extraction, while the second network is designed with expanded receptive fields specifically for global feature capture. Each network has differentiated quality characteristics matched to its extraction task, improving overall adaptability while preserving local precision
4Measurement precision
If traditional hand-designed features are used, then the computational cost is reduced, but the segmentation precision deteriorates due to limited feature expression ability
Solution Approach 1:
The patent replaces traditional hand-designed feature extraction methods with deep learning-based automatic feature extraction using convolutional neural networks. This substitution enables the system to automatically learn and extract relevant features from images, significantly improving segmentation precision while the parallel network architecture optimizes computational efficiency
Data Source
AI summary
The present invention discloses a method for segmenting pedestrians in roadside images using a variable-scale multi-feature fusion convolutional network. It addresses the challenge of significant changes in pedestrian scale by using two parallel convolutional neural networks to extract the local and global features at different scales, and then fusing them to obtain a variable-scale multi-feature fusion convolutional neural network, and this network is trained using roadside pedestrian images to realize accurate pedestrian segmentation, avoiding issues with boundary fuzziness and missing segments commonly found in single-network methods.