Autonomous Vehicle Position Detection Using Slender Linear Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deep learning algorithms fail to accurately detect linear objects in visual images, leading to inaccurate positioning of autonomous vehicles.
Innovation Solution
The method employs a slender convolution kernel neural network model with both underlying and high-level neural network layers to identify feature and size information of linear objects, which is then matched with preset coordinate system map information to determine the vehicle's position.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional deep learning algorithms are used to detect linear objects, then the detection process is simple, but the detection precision is insufficient for dense and slender linear objects
Solution Approach 1:
The patent segments the detection process into two distinct stages: an underlying neural network layer for initial feature extraction and a high-level neural network layer for precise size determination. This segmentation allows each layer to specialize in specific tasks, improving overall detection precision for slender linear objects while managing algorithmic complexity through functional division.
Solution Approach 2:
The patent introduces a dimensional transformation by converting the detection problem from direct image space analysis to a feature space representation through the underlying neural network layer, then further refining it in the high-level layer. This dimensional change enables better handling of dense and slender linear objects by representing them in a more suitable feature space.
2Measurement precision
If conventional deep learning algorithms are used, then the system is easier to implement, but the position determination accuracy is insufficient
Solution Approach 1:
The system is segmented into multiple processing stages: image acquisition, underlying neural network processing, high-level neural network processing, and coordinate system matching. Each stage produces intermediate results that are progressively refined, enabling accurate position determination while organizing system complexity into manageable modules.
Solution Approach 2:
The patent introduces intermediate representations (feature information from the underlying layer and size information from the high-level layer) that act as mediators between the raw image data and the final position determination. These intermediaries enable accurate positioning by providing structured information that bridges the gap between image input and position output.
3Reliability
If the linear object is dense and slender, then it provides good reference function for positioning, but it cannot be accurately detected by conventional algorithms
Solution Approach 1:
The detection of dense and slender linear objects is segmented into feature extraction by the underlying neural network layer and size determination by the high-level neural network layer. This segmentation allows the system to capture both the structural characteristics and dimensional properties of these challenging objects, improving detection accuracy while maintaining their value as reliable positioning references.
Solution Approach 2:
The patent transforms the detection of dense and slender objects from direct pixel-based measurement to feature-based representation through neural network layers. This dimensional change in the data representation enables accurate detection of objects that are difficult to measure directly in image space, preserving their reliability as positioning references.
Data Source
AI summary
The present application provides autonomous vehicle based position detection method and apparatus, a device and a medium, where the method includes: identifying an obtained first visual perception image according to an underlying neural network layer in a slender convolution kernel neural network model to determine feature information of the target linear object image, and identifying the feature information of the target linear object image by using a high-level neural network layer in the slender convolution kernel neural network model to determine size information of the target linear object image; further, matching the size information of the target linear object image with preset coordinate system map information to determine a position of the autonomous vehicle. Embodiments of the present application can accurately determine the position of the autonomous vehicle.