Rear Object Mapping Using 3D Ground and Virtual Planes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Agricultural and construction vehicles face challenges in navigating areas behind them due to limited visibility, which increases the risk of collisions with obstructions, especially when reversing.
Innovation Solution
The implementation of a vehicle navigation system that utilizes a set of cameras to capture images behind the vehicle, generating a disparity image and a 3D point cloud to identify a ground plane and virtual planes. This system applies virtual planes to the disparity image to create an object map, which highlights obstructions above a certain size and height, allowing the vehicle to navigate safely.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the vehicle uses sensors to detect objects behind it, then collision risk is reduced, but the precision of object detection and distance measurement deteriorates due to limited visibility and sensor constraints
Solution Approach 1:
The system transitions from 2D camera images to a 3D point cloud representation by introducing depth information through disparity calculation. This dimensional transformation enables precise distance measurement and spatial understanding of objects behind the vehicle, resolving the measurement precision limitation of traditional 2D sensors.
Solution Approach 2:
The system introduces virtual planes as intermediary structures between the ground plane and objects. These virtual planes serve as reference surfaces for height measurement and object classification, enabling more accurate detection and characterization of objects in the rear environment.
2Device complexity
If the vehicle navigates using basic sensor data, then the system complexity is low, but the ability to identify and classify objects by size and height deteriorates
Solution Approach 1:
The system segments the rear environment into distinct spatial zones using ground plane and virtual planes. This segmentation divides the complex 3D space into manageable regions (below virtual plane, between virtual planes, above virtual plane), enabling systematic object classification without requiring overly complex processing.
Solution Approach 2:
By transforming 2D image data into a 3D point cloud and introducing height dimension through virtual planes, the system enables object classification by size and height while maintaining reasonable computational complexity. The 3D representation provides natural height information that simplifies object categorization.
3Area of stationary object
If the vehicle uses multiple cameras to capture images, then the coverage area behind the vehicle is improved, but the complexity of image processing and disparity calculation increases
Solution Approach 1:
The system merges multiple camera images into a unified 3D point cloud representation. By combining the data from multiple cameras and processing them together through disparity calculation, the system achieves comprehensive rear coverage while streamlining the processing pipeline through integration rather than separate handling of each camera feed.
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
An agricultural or construction vehicle generates an object map to facilitate navigation. The agricultural or construction vehicle captures an image of the area behind the agricultural or construction vehicle, and generates a disparity image reflecting depth behind the agricultural or construction vehicle. The agricultural or construction vehicle generates and processes a 3D point cloud representation of the area behind the agricultural or construction vehicle to identify a ground plane. The agricultural or construction vehicle dynamically generates a virtual plane parallel to the ground plane based on the movement and position of the agricultural or construction vehicle. After applying the virtual plane to the disparity image, the agricultural or construction vehicle generates an object map identifying locations of objects greater than a threshold size and the height of the virtual plane. The agricultural or construction vehicle navigates through the area behind the agricultural or construction vehicle using the generated object map.