Surround-view stereo camera ground plane reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional vehicle imaging systems using a single downward-facing camera are less reliable in detecting objects next to a vehicle, as they rely on two-dimensional landmarks and pixel motion analysis, which can fail to accurately identify three-dimensional obstacles.
Innovation Solution
A surround-view system employing multiple downward-looking fisheye cameras to reconstruct the ground plane and compare images, identifying areas of disagreement to detect three-dimensional objects, with a processor generating a composite image and tracking the object's position by calculating ray intersections and employing modules for shape and wheel identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single downward-facing camera is used to detect objects, then the system complexity is reduced, but the detection reliability deteriorates
Solution Approach 1:
The system divides the detection task into multiple independent camera units, each capturing images from different locations around the vehicle. Multiple cameras segment the monitoring space into overlapping fields of view, allowing each camera to focus on specific zones while collectively providing comprehensive coverage that improves detection reliability without requiring a single complex camera system
Solution Approach 2:
The system merges images from multiple cameras through image merging modules that combine data from different viewpoints. By merging multiple two-dimensional images into a composite three-dimensional representation, the system achieves more reliable object detection than any single camera could provide, while maintaining manageable system complexity through modular architecture
2Device complexity
If two-dimensional landmarks are used for object detection, then the detection method is simplified, but the accuracy in identifying three-dimensional objects deteriorates
Solution Approach 1:
The system transitions from two-dimensional landmark detection to three-dimensional object identification by incorporating depth information through multiple camera viewpoints. By merging images from different angles and calculating spatial relationships between detected landmarks, the system reconstructs three-dimensional object positions and shapes, achieving accurate identification of three-dimensional obstacles while building upon simplified two-dimensional detection methods
3Device complexity
If pixel motion analysis is used to detect obstacles, then the detection process is simplified, but the reliability in detecting stationary or slow-moving objects deteriorates
Solution Approach 1:
The system combines multiple detection methods including pixel motion analysis with static image comparison and three-dimensional reconstruction. By merging results from motion-based detection with position-based landmark analysis from multiple cameras, the system maintains simplified detection processes while improving reliability for detecting both moving and stationary objects through complementary detection approaches
Data Source
Figure 1
Figure 2
Figure 3
AI summary
When detecting an object of interest, such as a bicyclist passing a truck, two downward looking cameras both capture images of the cyclist and detect the cyclist as a deviation from the flat ground plane. The ground plane is reconstructed using a homography (projection) matrix of each camera and compared. Where the camera images do not agree, the ground is not flat. The cyclist is located as the intersection of the rays extending to either end of the area of disagreement between the images.