Dump Body Position Specification Using Trained Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems struggle to accurately specify the position of a dump truck's dump body in images, especially when ruts are present on unpaved sites, as they often misidentify edges of the rut as part of the truck's edge.
Innovation Solution
An image processing system that includes a data acquisition unit and a position specifying unit using a trained model to accurately identify and specify the position of the dump truck's dump body by distinguishing it from surrounding edges, employing neural networks and stereo imaging to triangulate three-dimensional positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If edge extraction is used to specify the position of the dump body, then the position can be specified using simple image processing, but the presence of ruts causes incorrect position specification due to false edge detection
Solution Approach 1:
The patent replaces traditional mechanical edge extraction algorithms with a deep learning-based image recognition model. The model automatically identifies the dump body in the image without being affected by rut edges, achieving both simplicity in operation and high accuracy in position specification.
2Speed
If traditional image processing methods are used, then the processing speed is fast, but the accuracy of dump body position specification deteriorates in the presence of ruts
Solution Approach 1:
The patent substitutes traditional edge detection algorithms with a trained neural network model that can quickly and accurately identify the dump body. The model processes images in real-time while maintaining high accuracy even when ruts are present, achieving both speed and precision.
3Measurement precision
If a trained model is used to specify the position of the dump body, then the position specification accuracy is improved, but the device complexity increases due to model training requirements
Solution Approach 1:
The patent performs the complex model training process in advance, before actual use. The trained model is then deployed for real-time dump body position specification, where it operates efficiently without requiring additional training. This preliminary action separates the complex training phase from the simple inference phase.
Solution Approach 2:
The patent creates a trained model that copies the knowledge and patterns learned during the training phase. This copied knowledge is then applied repeatedly to new images without requiring the original training data or process, simplifying the operational system while maintaining high accuracy.
Data Source
AI summary
An image processing system includes a data acquisition unit and a position specifying unit. The data acquisition unit acquires a captured image showing a drop target of a work machine in which a transport object is dropped. The position specifying unit specifies a position of a predetermined part of the drop target shown in the captured image based on the captured image and a position specifying model. The position specifying model is a trained model, which outputs a position of a predetermined part of a drop target shown in an image when the image is input. A display device may display information regarding the position of the predetermined part of the drop target of the transport object specified by the image processing system.