Objective
To design and implement AI-powered automotive solutions that enhance driver safety, vehicle intelligence, and situational awareness through computer vision, deep learning, and signal processing technologies.
Solution & Approach
Requirement Analysis & Configuration
- Defined use cases and datasets for driver monitoring, traffic sign detection, and sound-based safety alerts.
- Conducted data collection, pre-processing, and augmentation to train high-performance deep learning models.
- Configured development environments for training, inference, and deployment on embedded hardware.
Network Management and Development
- Developed AI algorithms using Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) for real-time event detection.
- Integrated AI-based modules with ECU software for driver fatigue detection, object recognition, and sound classification.
- Enabled edge-to-cloud connectivity for data synchronization and continuous model improvement.
Memory and Bootloader Integration
- Optimized memory utilization for AI models deployed on embedded controllers.
- Supported bootloader configuration for secure firmware updates of AI inference modules.
- Ensured persistent storage for trained model parameters and runtime calibration data.
Security Enhancement
- Incorporated secure boot mechanisms and encrypted model updates to prevent tampering.
- Applied data privacy and integrity checks as per ISO 21434 and GDPR guidelines.
- Deployed AI model authentication and checksum validation during runtime.
Testing and Validation
- Conducted offline training validation using Python-based frameworks (TensorFlow, PyTorch).
- Executed real-time functional validation on hardware platforms such as Renesas R-Car and NXP S32V.
- Performed scenario-based testing to assess accuracy under varying environmental and lighting conditions.
Impact
- Achieved high-accuracy AI models (>90%) for real-time detection and driver assistance.
- Enhanced vehicle safety and situational awareness through proactive alerts.
- Reduced false-positive rates using optimized model training and adaptive filtering.
- Strengthened AVIN’s position in intelligent mobility and ADAS prototype development.
Tech Stack
- TensorFlow
- Python
- PyTorch
- OpenCV
- NumPy
- Renesas R-Car V3H
- NXP S32V
- Linux
- QNX
- ISO 26262
- ISO 21434
- CNN
- RNN
