Driver Face Recognition Using Synthetic Head-Region Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In car sharing environments, existing driver recognition systems face challenges in accurately identifying registered drivers due to varying interior designs of vehicles and the need for extensive and costly data sets, leading to overfitting and recognition issues when drivers wear glasses or hats.
Innovation Solution
A driver recognition system utilizing a balanced open data set and a small number of custom data sets, with a machine learning-based approach that includes object detection and driver matching models, extracts unique information from the driver's head region to perform automatic recognition, controlling vehicle settings based on pre-registered data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning network is used for driver recognition, then recognition capability is improved, but cost of constructing learning data set increases significantly
Solution Approach 1:
The patent uses image synthesis technology to generate virtual driver images by copying and transforming real driver images. Synthetic images are created through geometric transformations, lighting adjustments, and background replacements, enabling the system to generate large quantities of training data without additional physical data collection costs.
Solution Approach 2:
The patent applies parameter changes by modifying image parameters such as lighting conditions, head positions, facial expressions, and background environments in synthetic images. This allows the creation of diverse training data from limited real images, reducing the need for extensive physical data collection while maintaining recognition accuracy.
2Measurement precision
If extensive custom data sets are collected for each vehicle model, then recognition accuracy is improved, but data collection time and cost increase
Solution Approach 1:
The patent performs preliminary actions by pre-processing real driver images into standardized formats and synthesizing virtual images in advance. This pre-synthesized image library can be quickly adapted to different vehicle models without requiring time-consuming on-site data collection for each new vehicle type.
Solution Approach 2:
The patent creates a universal training data framework where synthetic images generated for one vehicle model can be adapted and reused across multiple vehicle models. The image synthesis process uses transferable parameters and transformations that work across different vehicle interiors, reducing the need for model-specific data collection.
3Measurement precision
If driver recognition system is optimized for specific vehicle interior, then recognition accuracy is improved, but adaptability to different vehicle models decreases
Solution Approach 1:
The patent uses parameter changes to adjust image characteristics such as background color, lighting intensity, and spatial coordinates to match different vehicle interiors. This allows the same recognition algorithm to adapt to various vehicle models by simply modifying image parameters rather than retraining the entire system.
Solution Approach 2:
The patent introduces image synthesis technology as an intermediary layer between real driver images and the recognition system. This intermediary process transforms real images into standardized synthetic images that can be universally processed by the recognition algorithm, decoupling the recognition system from specific vehicle interior characteristics.
4Adaptability or versatility
If more drivers are registered in car sharing environment, then service capability is improved, but recognition accuracy decreases due to overfitting
Solution Approach 1:
The patent generates multiple synthetic variations of each driver's image through copying and transforming the original images. This creates diverse training samples for each driver, preventing the system from overfitting to limited real images while maintaining the ability to recognize multiple drivers in car sharing environments.
Data Source
AI summary
A system and method are provided for automatically extracting image data of a driver's face when a driver sits in a driver's seat and then starting an engine of a vehicle, based on determining whether the extracted face matches a pre-registered face, and automatically controlling a vehicle environment based on the matched face data, thereby maximizing the convenience of using the vehicle.