Dilated Convolutional Neural Network Depth Map Generation for Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition technologies require expensive LiDAR sensors or complex stereo camera setups with high computational demands for generating depth maps, limiting their effectiveness and efficiency.
Innovation Solution
An apparatus and method utilizing a dilated convolutional neural network (DCNN) to generate depth maps from single images, eliminating the need for LiDAR sensors and improving object recognition rates by processing RGB and gray images through convolutional modules and concatenation processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR sensor is used to obtain high-resolution reflection map for object recognition, then object recognition accuracy is improved, but device cost and complexity increase
Solution Approach 1:
The patent creates a virtual depth map (copy of depth information) from a single RGB image using DCNN, replacing the need for physical LiDAR sensors. The depth map generator generates pseudo-depth information that mimics what LiDAR would provide, achieving similar recognition accuracy without expensive hardware
Solution Approach 2:
The patent replaces the mechanical/optical LiDAR system with a computational approach using DCNN and depth map generation. Instead of using laser beams and physical sensors to obtain depth information, the system uses neural network processing of RGB images to synthesize depth maps, substituting mechanical measurement with computational generation
2Loss of information
If stereo camera is used to generate depth map based on disparity information, then depth information is obtained, but calibration complexity and computational demand increase
Solution Approach 1:
The patent extracts depth information from a single RGB image using DCNN, eliminating the need for stereo camera pairs and their associated calibration processes. The depth map generator takes only one image as input and extracts all necessary depth information through neural network feature processing, removing the complexity of multi-camera synchronization and calibration
Solution Approach 2:
The system generates a virtual depth map that copies the essential depth information normally obtained from stereo disparity, but creates it computationally from a single image rather than through complex stereo processing and calibration procedures
3Measurement precision
If stereo camera with high-resolution images is used for object recognition, then recognition detail is improved, but computational complexity increases
Solution Approach 1:
The DCNN performs preliminary feature extraction and depth map generation from the RGB image before object recognition is attempted. By pre-processing the image to create a depth map with semantic information, the system reduces the computational burden on subsequent recognition stages, as the hard work of extracting meaningful features is already done
Data Source
AI summary
An apparatus for recognizing an object using an image includes a depth map generator that generates a depth map using a feature map of the image based on a dilated convolutional neural network (DCNN) and an object recognition device that recognizes the object using the depth map generated by the depth map generator and the image.