Vehicle Attitude Estimation from Point Clouds Without Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In unmanned electric shovel operations, accurately estimating the attitude of vehicles, such as mining trucks, is challenging due to the need for precise geometric information, which existing technologies often fail to provide with high accuracy and efficiency.
Innovation Solution
A vehicle attitude estimation method using point cloud data from Lidar scans to obtain target surfaces and corners, employing bounding box estimation and region growth algorithms to determine the vehicle's attitude, ensuring accurate positioning and orientation without relying on deep learning algorithms, thus reducing operational costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning algorithms are used for vehicle attitude estimation, then measurement precision may be improved, but device complexity and operational costs increase
Solution Approach 1:
The patent replaces expensive and complex deep learning algorithms with simpler, more efficient geometric feature extraction methods. By using basic geometric primitives (planes, cylinders, cones) and straightforward fitting algorithms, the system achieves accurate vehicle attitude estimation without the computational burden and high operational costs associated with deep learning models.
Solution Approach 2:
The patent substitutes complex computational mechanisms (deep learning) with simpler geometric and mathematical methods. The solution uses point cloud data processing combined with geometric feature detection and fitting algorithms to directly compute vehicle attitude, replacing the need for training and executing complex neural networks.
2Measurement precision
If deep learning algorithms are used for vehicle attitude estimation, then measurement precision may be improved, but operational costs increase
Solution Approach 1:
The patent employs computationally inexpensive algorithms that require minimal processing power and energy consumption. The geometric feature extraction and fitting methods used in the patent are significantly less resource-intensive than deep learning algorithms, thereby reducing operational costs while maintaining estimation accuracy.
Solution Approach 2:
The patent replaces energy-intensive deep learning computations with efficient geometric processing operations. By utilizing simple mathematical fitting and geometric reasoning on point cloud data, the system achieves accurate results with fraction of the computational energy required by deep learning approaches.
3Measurement precision
If geometric information is extracted from point cloud data, then vehicle attitude estimation accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent segments the complex task of vehicle attitude estimation into distinct geometric feature extraction steps. It identifies and extracts specific geometric features (planar surfaces, cylindrical bodies, conical structures) from the point cloud data separately, then combines these features to determine overall vehicle attitude. This segmentation simplifies the processing compared to applying a monolithic complex algorithm.
Solution Approach 2:
The patent applies different geometric fitting methods to different regions or features of the vehicle based on their local geometric characteristics. For example, it uses plane fitting for flat surfaces, cylinder fitting for rounded bodies, and cone fitting for tapered structures. This localized approach to feature extraction improves accuracy while keeping each individual processing step relatively simple.
Data Source
AI summary
Provided are a vehicle attitude estimation method, an electronic device and a storage medium, relates to a technical field of data processing, and in particular to fields of automatic driving, intelligent transportation, Internet of Things, big data and the like. A specific implementation solution includes: obtaining first target data, based on point cloud data of a vehicle, the first target data being capable of constituting a target surface of the vehicle; performing attitude estimation on a target body for surrounding the vehicle, based on the first target data, to obtain an estimation result; and estimating an attitude of the vehicle, based on the estimation result. According to the implementation solution, precise or accurate estimation of the attitude of the vehicle may be achieved.