Ground Plane Segmentation Network for Depth Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The inefficiency of image depth recognition due to the complexity and large number of network parameters in image segmentation networks, leading to slow processing times.
Innovation Solution
A deep recognition model training method that constructs a ground plane segmentation network by reducing the number of parameters in high-resolution networks, using a multi-stage architecture with adjusted channel counts and feature fusion, and generates a target height loss for the depth recognition network to improve segmentation speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If an image segmentation network with many parameters is used to perform depth recognition, then the accuracy of segmentation is improved, but the processing time increases and efficiency decreases
Solution Approach 1:
The patent changes the parameters of the segmentation network by replacing the original complex network architecture with a simplified network that has fewer parameters. Specifically, it uses a network with reduced channel dimensions and fewer layers while maintaining the core functionality of ground plane segmentation, thereby reducing processing time while preserving acceptable segmentation accuracy.
Solution Approach 2:
The patent extracts and removes unnecessary complex components from the segmentation network. It eliminates redundant convolutional layers and reduces the channel dimensions of feature maps, keeping only the essential elements needed for ground plane segmentation, thus simplifying the overall network structure and improving processing speed.
2Productivity
If the number of network parameters is reduced to improve processing speed, then the efficiency of depth recognition is improved, but the segmentation accuracy may deteriorate
Solution Approach 1:
The patent applies local quality by focusing computational resources on the most critical parts of the image for ground plane segmentation. Instead of uniformly processing all features with high complexity, it uses asymmetric channel dimensions and selective feature fusion to concentrate processing power on regions and features that most impact segmentation accuracy, thereby maintaining precision with reduced overall parameters.
Data Source
AI summary
A deep recognition model training method applied to an electronic device is provided. The method includes obtaining a ground plane area by segmenting a first image using a ground plane segmentation network. A projection image of the first image is generated based on the first image, an initial depth image corresponding to the first image, and a pose matrix. A target height loss of a depth recognition network is generated, and a depth loss of the depth recognition network is obtained according to a gradient loss between the initial depth image and the first image and a photometric loss between the projection image and the first image. A depth recognition model is obtained by adjusting the depth recognition network based on the depth loss and the target height loss.