Retaining Wall Detection Using LiDAR and Ground Elimination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting retaining walls and obstacles behind vehicles in mining areas face challenges such as reduced image quality due to dust, high costs and inefficiencies in deep learning-based laser point cloud algorithms, and issues with false detection and omission in traditional rule-based algorithms.
Innovation Solution
A method and system that uses a vehicle sensor system comprising a combinatorial navigation unit and a single-threaded laser radar unit to acquire data, which is then processed through filtering, dilation, coordinate transformation, clipping, and noisy point elimination to obtain processed data for detecting retaining walls. This system also determines ground points and calculates distance and integrity information of the retaining wall.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If image-based detection algorithms are used, then detection coverage is improved, but detection precision deteriorates due to dust and illumination conditions
Solution Approach 1:
The patent combines image data from cameras with laser point cloud data from LiDAR sensors to create a fused detection result. The image-based algorithm provides broad coverage while the laser-based algorithm provides precise distance measurement, compensating for the weaknesses of each individual modality in dusty mining environments
Solution Approach 2:
The patent introduces a data fusion module as an intermediary that processes and integrates information from both image sensors and laser sensors. This mediator reconciles the conflicting strengths and weaknesses of the two detection methods to achieve both coverage and precision
2Measurement precision
If point cloud-based deep learning algorithms are used, then detection precision is improved, but productivity deteriorates due to high time cost and economic cost
Solution Approach 1:
The patent segments the detection task into two parts: a rough detection phase using traditional rule-based algorithms for quick processing, and a refinement phase using deep learning only on uncertain or critical cases. This segmentation reduces the overall computational burden while maintaining precision where needed
Solution Approach 2:
Instead of applying computationally expensive deep learning algorithms to all detection cases, the patent applies them partially only to cases where rule-based algorithms are uncertain or where high precision is critical, achieving good precision with reduced processing time
3Productivity
If traditional rule-based algorithms are used, then productivity is improved, but detection precision deteriorates due to false detection and omission
Solution Approach 1:
The patent implements a feedback mechanism where the results from rule-based algorithms are evaluated, and cases with low confidence or high uncertainty are fed back to the deep learning algorithm for re-evaluation. This feedback loop corrects false detections and omissions while maintaining high processing efficiency
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system accurately detects distance information between the vehicle and retaining walls, ensuring safe backing operations, and provides integrity information of the retaining walls, thus preventing collisions with obstacles and ensuring driving safety.
Implementation Method 1
a single-threaded laser radar unit, the single-threaded laser radar unit is used for acquiring and outputting point cloud data related to the retaining wall behind the automatic driving vehicle
Data Source
AI summary
The invention discloses a method and system for detecting a retaining wall suitable for an automatic driving vehicle. The method comprises: obtaining original data acquired during a backing process of a vehicle and used for detecting a retaining wall, and processing the original data to obtain finally processed data of the retaining wall; obtaining ground data during the backing process of the vehicle, and sequentially performing ground determination and ground elimination on the ground data to obtain non-ground point cloud data; and obtaining distance information between a rear end of the vehicle and the retaining wall and integrity data of the retaining wall by calculation according to the finally processed data of the retaining wall and the non-ground point cloud data.