Smart Glasses Pose Tracking by Type-Specific Recognition Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pose recognition models for smart glasses in vehicles are complex and require extensive training due to the varied shapes and designs of smart glasses from different manufacturers, making them difficult to train and implement effectively.
Innovation Solution
A method involving a data-based glasses-type recognition model that identifies the specific shape of smart glasses, followed by selecting a corresponding pose recognition model trained for that type, simplifying the training process and reducing complexity by restricting the search area to a bounding box.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single pose recognition model is trained to recognize multiple glasses types from different manufacturers, then the model can handle diverse smart glasses shapes, but the training complexity and data requirements increase significantly
Solution Approach 1:
The patent divides the pose recognition task into two separate models: a glasses type recognition model that first identifies the specific glasses type, and a pose recognition model that then determines the pose based on the recognized type. This segmentation allows each model to be specialized and simpler, rather than requiring one complex model to handle all glasses types simultaneously.
Solution Approach 2:
The glasses type recognition model serves as an intermediary between the camera input and the pose recognition model. It classifies the glasses type first, and this classification acts as a mediator that enables the pose recognition model to be selected or configured appropriately for that specific type, reducing the overall complexity.
2Adaptability or versatility
If a single pose recognition model is trained to recognize multiple glasses types, then it can determine poses for various manufacturers' glasses, but the training data sets required become extensive and difficult to obtain
Solution Approach 1:
By segmenting the recognition task into type identification and pose determination, the patent reduces the training data requirements for each individual model. The type recognition model needs data to distinguish between glass types, while the pose recognition model needs data for accurate pose estimation of a specific type, rather than requiring exhaustive data covering all possible variations in a single model.
Solution Approach 2:
The glasses type recognition model performs preliminary classification before pose recognition. This preliminary action allows the system to narrow down the scope of the pose recognition task to a specific glasses type, thereby reducing the amount of training data needed for the pose recognition model compared to training it on all possible glasses types simultaneously.
3Reliability
If the search area for pose recognition covers the entire camera image, then all glasses poses can be detected, but the recognition time and computational latency increase
Solution Approach 1:
The patent extracts the glasses region from the entire camera image by using the glasses type recognition model to first identify and locate the glasses. This extraction creates a focused search area that contains only the relevant glasses region, eliminating the need to process the entire image for pose recognition and thereby reducing computational latency.
Solution Approach 2:
By applying pose recognition specifically to the extracted glasses region rather than the entire image, the patent implements local quality processing. The computational resources are concentrated on the local area where the glasses are located, maintaining detection reliability while significantly reducing the processing time compared to analyzing the full image.
Data Source
AI summary
A method for determining a pose of a pair of smart glasses in an interior of a mobile machine includes capturing a camera image of an interior of the mobile machine and determining a relevant glasses type for the pair of smart glasses imaged by evaluating a data-based glasses-type recognition model using the captured camera image. The glasses-type recognition model is trained to assign a camera image to a relevant glasses type of one or more pairs of smart glasses. One of a plurality of data-based pose recognition models is selected depending on the recognized glasses type. The pose recognition models are each trained for one glasses type to determine a glasses pose of the corresponding pair of smart glasses according to the camera image. The glasses pose of the pair of smart glasses of the corresponding glasses type is determined via the selected pose recognition model.