Autonomous Vehicle Depth Estimation Using Terrain Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles, particularly semi-trailer trucks, face limitations in long-distance perception, as existing systems can only detect objects within a limited range (up to 300 meters), which may not be sufficient for safe maneuvering and proactive collision avoidance.
Innovation Solution
The implementation of an image processing method that uses cameras to determine the 3D position of objects by receiving bounding box information, calculating camera coordinates, and utilizing terrain maps to assign world coordinates, allowing for the detection of objects up to 1000 meters away, enhancing the vehicle's perception and safety features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Length of stationary object
If existing object detection systems are used in autonomous vehicles, then the system complexity remains manageable, but the detection range is limited to 300 meters which is insufficient for safe maneuvering
Solution Approach 1:
The patent introduces terrain maps as an intermediary data structure to bridge the gap between 2D image coordinates and 3D world coordinates. The terrain map stores pre-computed elevation and spatial information, allowing the system to extend detection range without proportionally increasing processing complexity. By using the terrain map as a mediator, the system can perform depth estimation for objects up to 1000 meters away while maintaining manageable computational requirements.
Solution Approach 2:
The patent transitions from 2D image plane coordinates to 3D world coordinates by introducing the depth dimension through terrain map data. This dimensional transformation allows the system to leverage existing 2D camera systems for 3D spatial understanding, extending detection range without requiring additional sensors or dramatically increasing system complexity. The bounding box information is enhanced with depth estimates derived from terrain elevation data.
2Reliability
If the detection range is extended to 1000 meters using the image processing method, then collision avoidance capability improves, but the computational processing requirements increase
Solution Approach 1:
The patent performs preliminary computation of terrain maps offline or in advance, storing pre-computed elevation and spatial relationship data. During real-time operation, the system only needs to query the pre-computed terrain map with object coordinates to obtain depth estimates, rather than performing complex 3D reconstruction calculations. This preliminary action significantly reduces real-time computational requirements while maintaining extended detection range and high reliability for collision avoidance.
3Adaptability or versatility
If bounding box information is used for depth estimation, then the method can work with standard object detection outputs, but the precision of depth estimation decreases compared to specialized sensors
Solution Approach 1:
The patent changes the parameters used for depth estimation by incorporating terrain map elevation data alongside bounding box coordinates. Instead of relying solely on 2D image geometry, the system queries the terrain map for elevation information at the projected 3D location and uses this additional parameter to compute depth. This parameter enhancement maintains compatibility with standard object detection outputs while improving depth estimation precision through the addition of terrain-based spatial information.
Data Source
AI summary
Image processing techniques are described to receive bounding box information that describes a bounding box located around a detected object in an image, determine one or more positions of one or more reference points on the bounding box, determine, for each reference point, 3D world coordinates of a point of intersection of the reference point and the road surface, and assign the 3D world coordinates of the one or more reference points to a location of the detected object.