ADAS Engineering Services
Advanced Driver Assistance Systems (ADAS) and autonomous driving are revolutionizing the automotive industry, enhancing vehicle safety, efficiency, and driving experiences. The future of ADAS is driven by AI, sensor fusion, augmented reality, and over-the-air (OTA) updates, ensuring continuous improvement and enhanced road safety.
Arrow Automotive CoE offers comprehensive ADAS/AD solutions, including integration of advanced sensors (Camera, Radar, Lidar), optimization & porting of ADAS algorithms, Autosar-based ECU development, driver monitoring systems, AI-based data annotation, model-based development, validation, and testing of various ADAS/AD functions.
Our Experience with ADAS Applications
Key Service Offerings
Algorithm Optimization/Porting
- Algorithms optimization and porting on target platforms for object detection and pedestrian detection
Camera/Vision based Solutions
- Camera model, algorithms, ML model training & porting
- Image/video tuning, video analytics, lens distortion correction
Data Sourcing and Annotation
- Image/video data for algorithm training, 2D/3D labelling (bounding box, cuboids) and semantic segmentation
Sensor & Data Fusion
- Sensor experience – Camera, Radar, Lidar, Ultrasonic etc.
- Fusion of multi sensor inputs for actuation/warning
Connectivity and Pre-processing
- Connectivity and drivers for a variety of platforms
- Performance optimization for low memory DSP/GPU/ISPs
Compute Platforms
- Heterogeneous platform-based execution on multi-core/OS/SoC
Case Studies
Pedestrian and Object Detection Algorithm Optimization and Porting on Hydra DSP
- Optimized & Ported Algorithms – Successfully optimized and ported proprietary Object Detection (OD) and Pedestrian Detection (PD) algorithms for constrained memory environments.
- Memory & Performance Optimization – Achieved ~30fps at 720×480 resolution while fitting execution for four ADAS algorithms within 32 KB instruction memory & 8 KB data memory using memory overlays.
- Algorithm Adaptation & Integration – Ported and optimized Camera Model (Fish Eye to Pinhole ~5K LOC), OD (~110K LOC), and PD with KCF Tracker (~80K LOC) for efficient ADAS processing.
- Efficient Code Implementation – Converted C++ PC-based floating-point code with trigonometric operations into optimized embedded C with overlays, ensuring seamless execution on Hydra DSP.
ADAS Algorithm Porting – Side View Mirror Monitoring
- Optimized ADAS Algorithm Porting – Successfully ported Optical Self Diagnosis (OSD) and Camera Calibration algorithms for side view mirror monitoring on the Broadcom 89107 & Hydra DSP platform.
- Memory & Performance Optimization – Achieved ~30fps at 720×480 resolution while efficiently utilizing only 2 out of 8 DSP cores, despite low memory availability (32 KB instruction, 8 KB data memory).
- Enhanced Functionalities – Developed blind spot monitoring, glare recognition, and soil detection features to improve driving safety.
- Efficient Data Handling & Processing – Implemented floating-to-fixed point conversion and optimized DMA operations for seamless processing between ARM and DSP without shared memory.
ADAS Feature Enhancement & Algorithm Validation
- Enhanced ADAS Capabilities – Improved Adaptive Cruise Control and Emergency Brake Assist by integrating RADAR and Camera sensor fusion for enhanced situational awareness.
- Advanced Simulation & Testing – Developed 3D environmental modeling to simulate diverse conditions like rain, fog, and snow, ensuring robust algorithm validation and feasibility studies.
- ISO26262 & Automotive SPICE Compliance – Ensured safety and reliability by following MISRA coding guidelines, rigorous unit and integration testing, and field test support.
- Optimized Performance & Reusability – Delivered modular and reusable code, improving test coverage and reducing time to market through simulation-based validation.
AI-ML based Driver Behavior Assessment & Monitoring
- AI-Based Driver Monitoring – Developed a model to detect 9 driver behaviors (e.g., texting, eating, talking on the phone) for real-time distraction assessment and alert generation.
- High-Accuracy Deep Learning Model – Applied transfer learning on VGGNet & ResNet with a 35K+ image dataset, achieving 85% accuracy and optimizing for edge deployment using SqueezeNet.
- Efficient Edge Deployment – Successfully ported server-trained models to cabin camera devices (227×227 @ 30 FPS) using TensorRT with DeepStream, ensuring real-time processing.
- Improved Fleet Safety & Cost Savings – Enabled live monitoring of 1000+ heavy vehicles, reducing driver distractions by 50%, lowering accident rates, and cutting insurance premiums.
LIDAR-Based Collision Awareness System
- Advanced Collision Prevention – Developed a wired and wireless LIDAR-based system for real-time 360-degree environment monitoring, ensuring factory floor safety.
- End-to-End System Development – Conducted feasibility studies, hardware design, embedded software development, and component selection, integrating LIDAR sensors with haptic feedback.
- Enhanced Workplace Safety – Enabled real-time alerts to drivers, significantly reducing hazardous incidents and improving mobility for employees and vehicles.
- Cost & Risk Reduction – Minimized litigation risks and accident-related compensation costs, creating a safer and more efficient manufacturing environment.
Recently published insights
Awards & Recognitions
Next-Gen EV Charging Station Development
Add ISG Digital Case Study Award received for PTI Digital Platform Development
Start a conversation today
We invite you to join us for a 30-minute introductory call to explore our ADAS engineering services and hear about our success stories. This call will provide a valuable opportunity to understand how we can meet your needs and demonstrate the impact we deliver. Please fill out the form to secure your spot, and we look forward to connecting with you soon!
