Camera Distance Estimation Using Lane Plane and Depth Map
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining the distance between a vehicle and an object in front of it are limited by the use of LiDAR sensors or stereo cameras, which can be aesthetically unpleasing, less utilized due to performance issues, and prone to inaccuracies with objects having light reflection properties. Additionally, deep learning-based methods only provide relative information in 2D images.
Innovation Solution
A distance determination method and device that uses a camera to generate a target image, a lane plane image, and a depth map. The method involves setting a vanishing point, extracting line segments and lane lines, generating the lane plane image, and determining the target distance based on the lane plane image and depth map.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR sensor or stereo camera is used to obtain depth map, then distance measurement capability is improved, but device complexity and aesthetic appearance are worsened
Solution Approach 1:
The patent uses a standard 2D camera to capture images and generates a depth map through image processing algorithms rather than using specialized depth-sensing hardware like LiDAR or stereo cameras. This creates a functional copy of depth perception capability using simpler components, resolving the contradiction between measurement precision and device complexity
Solution Approach 2:
The patent replaces mechanical/optical depth sensing systems (LiDAR, stereo cameras) with a computational approach using 2D image processing and geometric calculations. The vanishing point-based lane plane generation and depth map estimation substitute physical depth-sensing mechanisms with algorithmic processing, eliminating the need for complex hardware while maintaining distance measurement capability
2Measurement precision
If LiDAR sensor or stereo camera is used to obtain depth map, then distance measurement capability is improved, but reliability is worsened due to performance issues and light reflection interference
Solution Approach 1:
The patent creates a robust depth estimation system that works reliably across different lighting conditions by using 2D image data and geometric processing rather than optical depth sensing that is sensitive to light reflection. The vanishing point-based approach provides consistent performance regardless of object reflectivity properties
Solution Approach 2:
The system uses the 2D camera's inherent capabilities to generate depth information through self-contained image processing algorithms. By leveraging the camera's own data and the geometric properties of lane markings, the system achieves reliable depth estimation without being affected by external factors like light reflection from objects
3Productivity
If deep learning-based depth map estimation is used, then processing speed is improved, but measurement precision is worsened by providing only relative information
Solution Approach 1:
The patent transforms the depth map from providing only relative depth information to providing absolute distance measurements by introducing the vanishing point parameter and lane plane geometric model. This parameter change enables the system to calculate actual distances while maintaining the efficiency of 2D image processing
Solution Approach 2:
The patent adds a geometric dimension to the 2D image processing by incorporating the vanishing point and lane plane concepts. This dimensional enhancement allows the system to derive absolute distance information from 2D images without requiring complex 3D sensing hardware or slow deep learning processing
Data Source
AI summary
A distance determination method includes generating a target image using a camera, generating a lane plane image based on the target image, generating a depth map based on the target image, and determining a target distance between the camera and a point in a real world corresponding to a target pixel in the lane plane image based on the lane plane image and the depth map.