Depth Map Generation from 2D Images via CNN Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition systems, especially in applications like autonomous driving, face challenges in determining depth information from 2D images due to the high computational requirements and the need for additional costly components like LiDAR scanners.
Innovation Solution
A system and method using convolutional neural networks (CNNs) that apply convolution and pooling operations to extract feature maps and subsequent deconvolution operations to generate depth maps from 2D images, eliminating the need for separate distance measuring devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR scanners are used to measure distance and spatial relationship, then depth information measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent creates a virtual depth map by copying and processing 2D image data through convolutional neural networks, rather than using physical LiDAR sensors. The system generates a depth representation (copy) from the existing 2D image, eliminating the need for separate distance measuring devices while achieving depth information extraction
Solution Approach 2:
The patent replaces the mechanical LiDAR scanning system with a computational approach using convolutional neural networks. Instead of using physical laser scanners to measure depth, the system uses software-based image processing with convolution operations, pooling, and deconvolution to extract depth information from 2D images
2Measurement precision
If fully connected neural networks are used to process high-resolution images, then object recognition accuracy is improved, but computation capacity requirements increase
Solution Approach 1:
The patent segments the image processing task into multiple convolutional layers, each handling local feature extraction. Instead of processing the entire high-resolution image at once with a fully connected network, the convolutional network divides the image into regions and processes them through multiple specialized layers, reducing overall computational burden while maintaining accuracy
Solution Approach 2:
The patent transforms the 2D image data into a 3D feature map through convolutional operations, adding a depth dimension to the feature representation. This dimensional transformation allows the network to capture spatial relationships more efficiently without requiring excessive computation, as the 3D structure encodes depth and hierarchy information
Data Source
AI summary
Embodiments of the disclosure provide systems and methods for determining depth information in a two-dimensional (2D) image. An exemplary system may include a processor and a non-transitory memory storing instructions that, when executed by the processor, cause the system to perform the various operations. The operations may include receiving a first feature map based on the 2D image and applying an extraction network having a convolution operation and a pooling operation to the first feature map to obtain a second feature map. The operations may also include applying a reconstruction network having a deconvolution operation to the second feature map to obtain a depth map.