Multi-Camera Object Detection for Fisheye Distortion Without Cropping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fisheye lenses used in autonomous driving suffer from large distortion, leading to loss of field of view and feature information when de-distortion is applied, and mapping to spherical or cylindrical images results in partial cropping, losing valuable image features.
Innovation Solution
A method and apparatus that utilize multiple cameras, including fisheye lenses, to capture images, determine high-dimensional parameter features, and fuse these features using a target object detection model, maintaining the order of cameras to enhance accuracy and robustness by converting and encoding images in a high-dimensional space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If de-distortion is applied to fisheye lens images, then the distortion is corrected, but the field of view and feature information are lost
Solution Approach 1:
The patent transforms 2D fisheye lens images into 3D spatial features by determining high-dimensional parameter features (position, orientation, scale) of camera parameters. This dimensional transformation allows the system to preserve the original field of view and feature information while achieving accurate distortion correction through geometric relationship analysis in the high-dimensional space.
Solution Approach 2:
The patent segments the fisheye lens image processing into independent modules: image acquisition, high-dimensional parameter feature determination, and feature fusion. By processing different aspects of the image separately and then fusing the results, the system maintains the original image information while correcting distortion through the high-dimensional parameter analysis.
2Ease of operation
If mapping to spherical or cylindrical images is applied, then the fisheye lens distortion is handled, but partial cropping occurs and valuable image features are lost
Solution Approach 1:
Instead of transforming the fisheye lens image into spherical or cylindrical format (which causes cropping), the patent creates a virtual copy of the image in high-dimensional space. By determining high-dimensional parameter features and fusing them with the original image features, the system preserves the complete original image information while handling the distortion through the high-dimensional parameter representation.
3Measurement precision
If multiple cameras are used to capture images, then the field of view and detection accuracy are improved, but the system complexity increases
Solution Approach 1:
The patent creates a universal processing framework that handles multiple camera inputs through a single high-dimensional parameter feature determination module and feature fusion mechanism. This multi-functional approach allows the system to process images from multiple cameras with different configurations using the same standardized procedure, reducing the complexity increase that would otherwise result from handling each camera separately.
Data Source
AI summary
A target object detection method, including: obtaining images collected by more than one camera installed on a target vehicle; determining a high-dimensional parameter feature in a high-dimensional space corresponding to parameter information of each camera; and fusing features of the images via a target object detection model according to the high-dimensional parameter features, and determining position information of a target object based on the fused features, an order of the cameras corresponding to the images being the same as an order of the cameras corresponding to the high-dimensional parameter features.