Monocular Depth Estimation via Point Cloud Matching
Solution Overview
Problem
Current methods for obtaining depth images, such as binocular vision, structured light, and time-of-flight cameras, are not accurate enough, which affects the safety of autonomous driving.
Innovation Solution
A method using a monocular camera connected to an electronic device that employs a deep learning network model to process images and estimate depth maps, followed by point cloud conversion and matching to improve depth image accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If binocular vision, structured light, or time-of-flight cameras are used to obtain depth images, then depth image acquisition is achieved, but the accuracy of depth images is insufficient
Solution Approach 1:
The patent replaces traditional mechanical/optical depth acquisition methods (binocular vision, structured light, time-of-flight) with a computational approach using monocular camera images and deep learning neural networks to estimate depth maps, thereby improving accuracy without additional hardware complexity
Solution Approach 2:
The patent changes the fundamental parameter of depth acquisition from physical/optical measurements to computational estimation based on image data and learned patterns from neural networks, enabling higher precision depth image generation
2Measurement precision
If traditional depth image methods are used, then system complexity is kept low, but depth image accuracy is not sufficient for safe autonomous driving
Solution Approach 1:
The patent substitutes complex hardware systems (multiple cameras, lasers, sensors) with a single monocular camera combined with computational algorithms and neural networks, reducing physical system complexity while improving measurement precision through software-based depth estimation
Data Source
AI summary
A method for obtaining depth images for improved driving safety applied in an electronic device processes first and immediately-following second images which have been captured and processes each to obtain two sets of predicted depth maps. The electronic device determines a transformation matrix of a camera between first and second images and converts the first predicted depth maps into first point cloud maps, and second predicted depth maps into second point cloud maps. The first point cloud maps are converted into third point cloud maps, and the second point cloud maps into fourth point cloud maps. First and fourth point cloud maps are matched and first and second error values are calculated, thereby obtaining a target deep learning network model. Images to be detected are input into the target deep learning network model and depth images are obtained.