Case-study - In-Vehicle Data Acquisition & Recording Platform for ADAS-Training
Project Brief
A Japanese OEM approached us to build and deploy an embedded in-vehicle data collection system that integrates with vehicle networks, sensors, cameras, GPS, and an HMI interface to collect real-world driving datasets for ADAS training across 5,000 km of Indian roads. This case study covers our design approach, technology stack, and key highlights in developing this data-acquisition system.
Requirement
The solution needed to be deployed in their existing vehicles while ensuring reliable vehicle connectivity, secure mounting, and an intuitive user interface. The customer required a cost-effective platform comprising:
Vehicle Computer: A compact and reliable vehicle computer that could be mounted in the vehicle and interface with the vehicle’s OBD-II, inclination sensor, road-view camera, driver-view camera, a push button, and an Android tablet for the user interface.
Embedded Datalogging Software: The software running on the vehicle computer should log CAN data, vehicle inclination, GPS location, road-view and driver-view camera feeds, traffic data, and push-button status in a time-synchronized manner and organize the collected data into 10-minute segments.
Android HMI: The system should interface with an Android device to control data logging and display the map/route, logging status of each peripheral, live feeds from the road-view and driver-view cameras, and USB storage status.
Design Solution
System Architecture
The diagram below shows the architecture of the data acquisition system in the vehicle.
Technology Stack Used
-
Qt
-
Android
-
Linux
-
Python
-
OpenStreetMap
-
GStreamer
Engineering Highlights
-
Custom Onboard Computer: An onboard computer with a rugged mechanical mount was designed and built to collect multiple data from the vehicle in Indian driving conditions.
-
Multiple Data Synchronization: Multiple data streams like CAN, GPS, Inclination sensor, camera videos, weather, and driver interactions were synchronized at specific intervals for an accurate training dataset.
-
Map & Navigation: A regional offline map is self-hosted on the onboard computer with OpenStreetMap for efficient rendering. Google My Maps is integrated for Routes API cost reduction.
-
Traffic Context Integration: Implemented an algorithm to track the current leg of the predefined route using live GPS location and retrieve traffic information using Google Routes API.
-
Camera Streaming Architecture: Designed Camera Streaming Architecture with GStreamer and RTSP for live streaming and logging of two cameras simultaneously.
Project Outcome
The customer was able to successfully record 5000 KMs of data from the test vehicle for their analysis and model training. We acquired two distinct business advantages as a result,
-
An Onboard Computer with vehicle integration and a dashboard that aided in real-time data collection.
-
A framework for vehicle data acquisition that supports the training of AI models for driver-assistance features.
If you are working on a vehicle data acquisition system, an ADAS training pipeline, or any embedded system that needs to perform reliably in demanding field conditions, our highly experienced engineering team is ready to help you with design and implementation. For any queries, kindly reach out to sales@zilogic.com