Neural Network Position Tracking for Mixed Reality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing mixed reality and augmented reality devices face challenges in accurately determining the position of input devices in 3D space, leading to unnatural interactions due to offsets between real and virtual environments.
Innovation Solution
A method involving an industrial robot with a light-emitting element, capturing images with multiple cameras, and training a fully connected artificial neural network to determine the position of the light-emitting element or input device, using pixel coordinates from CCD sensors as input, and dividing the camera field of view into coordinate areas to reduce processing time and latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image processing and image recognition methods are used to determine the position of the light-emitting element, then the system can identify the object's location, but the computing time is excessive and training speed is reduced
Solution Approach 1:
The patent extracts only the essential information needed for position determination - specifically the coordinates of illuminated pixels from camera sensors - and feeds this directly into the neural network. This extraction approach bypasses time-consuming image processing and recognition steps, achieving both accurate position determination and reduced computing time.
Solution Approach 2:
The patent replaces traditional mechanical image processing systems with a neural network-based system that directly processes pixel coordinate data. This substitution eliminates the need for complex image recognition algorithms while maintaining position determination accuracy and significantly reducing processing time.
2Area of stationary object
If the camera field of view is processed as a single unit by the neural network, then the system maintains comprehensive coverage, but processing time and latency increase
Solution Approach 1:
The patent divides the camera sensor coordinates into multiple ranges and assigns different coordinate ranges to different subnetworks. This segmentation allows parallel processing of different spatial regions, reducing overall processing time and latency while maintaining comprehensive field of view coverage.
Solution Approach 2:
The patent introduces a new dimension of processing by organizing the neural network into multiple subnetworks that operate in parallel on different coordinate ranges. This dimensional organization of processing units enables simultaneous computation across the entire field of view, effectively reducing processing time.
3Reliability
If a single artificial neural network processes all coordinate data, then the system maintains unified processing, but training time and computational resources are excessive
Solution Approach 1:
The patent segments the single neural network into multiple subnetworks, each responsible for processing specific coordinate ranges. This segmentation reduces the computational burden on each individual network, enabling faster and more efficient training while maintaining reliable position determination through coordinated processing of all subnetworks.
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
This approach enables precise and reliable determination of the input device's position, improving interaction accuracy and reducing latency, allowing for more intuitive engagement with virtual content in mixed reality environments.
Implementation Method 1
Any position in the workspace of a robot, in particular an industrial robot, is approached, with a light-emitting element attached to the robot
Implementation Method 2
coordinates of the pixels of the sensors of the cameras illuminated by the light-emitting element being determined
Data Source
Figure 1~2
Figure 3~3b
Figure 4~4b
AI summary
The invention relates to a method for training an artificial neural network for determining the position of an object. Any position in the working area of a robot having a light-emitting element is approached. At least two images of the object are captured by cameras (24). Coordinates of the object are determined and used as input parameters for the input layer of the artificial neural network. The invention further relates to a device for training an artificial neural network and a method and a system for determining a position of an object.