Vehicle Pose Calculation Using State-Based Camera and LiDAR Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In manufacturing environments where vehicles undergo various manufacturing steps, the changing appearance of vehicles complicates accurate position and orientation calculation using sensors, leading to decreased detection accuracy and increased costs due to the need for multiple sensor types and extensive preprocessing.
Innovation Solution
A calculation device and system that utilize a combination of sensors (camera and LiDAR) to classify vehicle states based on appearance, allowing for reduced preprocessing and cost by using the most suitable sensor for each state, enabling efficient position and orientation calculation without extensive learning processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning model is trained to detect the appearance shape of the moving object, then the detection accuracy is improved, but the learning time and preprocessing load increase significantly
Solution Approach 1:
The patent segments the detection task by creating multiple specialized machine learning models, each trained on data from a specific manufacturing step. This allows the system to use a lightweight, step-specific model rather than one large comprehensive model, reducing overall preprocessing time while maintaining accuracy for each specific state.
Solution Approach 2:
The patent performs preliminary classification to determine the manufacturing step of the moving object before selecting and applying the corresponding pre-trained machine learning model. This preliminary action avoids the need to train a new comprehensive model, as the appropriate specialized model is already prepared in advance for each manufacturing step.
2Device complexity
If LiDAR is used to detect the moving object without preprocessing, then the preprocessing load is reduced, but the cost increases
Solution Approach 1:
The patent makes the camera system multi-functional by training it to perform detection tasks equivalent to LiDAR through machine learning. The camera, being a lower-cost sensor, is enhanced with AI capabilities to extract position and orientation information without requiring expensive LiDAR hardware, thus reducing cost while maintaining functional capability.
Solution Approach 2:
The patent replaces the mechanical/optical LiDAR system with a camera-based visual system enhanced by machine learning algorithms. This substitution uses computational processing instead of specialized hardware to achieve similar detection functionality, reducing the need for expensive sensors while maintaining detection capability.
3Measurement precision
If multiple machine learning models are trained for each appearance state, then the detection accuracy for each state is improved, but the total preprocessing load and cost increase
Solution Approach 1:
The patent segments the detection task by creating multiple specialized machine learning models, each trained on data from a specific manufacturing step. This allows the system to use a lightweight, step-specific model rather than one large comprehensive model, reducing overall preprocessing time while maintaining accuracy for each specific state.
Solution Approach 2:
The patent performs preliminary classification to determine the manufacturing step of the moving object before selecting and applying the corresponding pre-trained machine learning model. This preliminary action avoids the need to train a new comprehensive model, as the appropriate specialized model is already prepared in advance for each manufacturing step.
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 reduces the load and cost associated with preprocessing while maintaining accurate position and orientation calculations by dynamically selecting the appropriate sensor type based on the vehicle's state, thereby enhancing detection accuracy and reducing the need for extensive data preparation and learning times.
Implementation Method 1
acquire second sensor information that is output from a second sensor of a different kind from the first sensor by detecting the moving object classified into a second state different from the first state among the states via the second sensor from an outside of the moving object
Implementation Method 2
in a case where the appearance shape of the moving object included in measurement point group data output from a light detection and ranging (LiDAR) is detected
Data Source
AI summary
A calculation device includes a sensor information acquisition unit configured to acquire first sensor information acquired by a first sensor that detects a moving object classified into a first state from an outside of the moving object, and acquire second sensor information acquired by a second sensor that detects the moving object classified into a second state from the outside, and a calculation unit configured to calculate at least one of a position and an orientation of the moving object by using the first sensor information without executing preprocessing, and execute the preprocessing to calculate at least one of the position and the orientation of the moving object by using the second sensor information.