Monocular Image Distance Estimation Without Camera Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing distance determination techniques using vehicle sensors are highly sensitive to camera calibration and placement, limiting their accuracy and requiring additional sensors like radar or lidar, which increases complexity and cost.
Innovation Solution
A method using monocular image data from a camera to generate feature and depth maps, classify objects, and estimate distances through a series of machine-learning programs, independent of camera calibration, allowing for improved accuracy and potential fusion with other sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera calibration and placement are used for distance determination, then distance can be estimated, but accuracy is highly sensitive to calibration errors and placement variations
Solution Approach 1:
The patent replaces traditional geometric/calibration-based distance determination with a machine learning-based approach. Instead of relying on precise camera calibration and geometric models, the system uses neural networks trained on image data to directly predict distance, eliminating sensitivity to calibration errors and placement variations.
Solution Approach 2:
The patent changes the fundamental parameters used for distance determination from calibration-dependent geometric parameters to calibration-independent features extracted by machine learning models. The system transforms input images through multiple processing stages, extracting increasingly abstract features that are inherently invariant to camera calibration and placement.
2Measurement precision
If additional sensors like radar or lidar are added, then distance determination accuracy improves, but device complexity and cost increase
Solution Approach 1:
The patent makes the camera system multi-functional by enabling it to perform both traditional image processing tasks and accurate distance determination through machine learning. The same camera hardware and image data are used for multiple purposes, eliminating the need for dedicated distance measurement sensors like radar or lidar.
Solution Approach 2:
The patent creates a virtual depth map as a computational copy of the physical depth information. Instead of using physical sensors to directly measure distance, the system generates a depth map from image data that can be processed and fused with other sensor data, providing accurate distance information through software rather than additional hardware.
3Device complexity
If traditional distance determination methods are used, then system simplicity is maintained, but accuracy is limited and additional sensors are required
Solution Approach 1:
The patent adds a computational dimension to the image processing pipeline by introducing multiple machine learning processing stages. The system transforms 2D image data through feature extraction, feature map generation, and depth map creation, effectively adding processing dimensions that enable accurate distance determination from monocular images without additional sensors.
Data Source
AI summary
A computer includes a processor and a memory, and the memory stores instructions executable by the processor to receive an image frame from a camera, generate a feature map from the image frame, generate a depth map from the feature map, classify an object in the image frame based on the feature map, and estimate a distance to the object based on the depth map and based on an input to generating the feature map.