Vehicle Telemetry using Raspberry Pi with Kafka-based Anomaly Detection

In 2019, I built a hobby project to collect real-time vehicle parameters from the OBD-II port using an ELM327 Bluetooth adapter and a Raspberry Pi. This setup allowed me to display vehicle speed and RPM on a custom instrument console. As a demonstration, the Raspberry Pi connected to the ELM327 via Bluetooth and parsed live vehicle data directly from the CAN bus.

The hardware flow looked like this:

• Vehicle → ELM327 / CAN Bus
• ELM327 → Bluetooth → Raspberry Pi
• Raspberry Pi → Instrument Console (Python-based UI)

I have now planned to enhance the project with real-time streaming and anomaly detection:

• The Raspberry Pi connects to a Kafka broker using a 4G modem.
• Vehicle telemetry data such as speed, RPM, coolant temperature, and throttle position is pushed to Kafka topics.
• These streams are processed using Apache Spark, and anomalies are detected with a trained ML model.

✅ Implemented Components:

• ELM327 Bluetooth integration
• Real-time data collection on Raspberry Pi
• Custom instrument display for speed and RPM

🔜 Planned Enhancements:

• Kafka streaming pipeline with 4G modem
• Apache Spark for telemetry processing
• ML-based failure detection using vehicle health indicators

Demo & Source Code

You can check out the working video demo and the source code in the GitHub repository:

Github Repo

💻 GitHub Repo

Tech Stack

• Raspberry Pi 3B+
• Python 3
• ELM327 OBD-II Adapter
• Apache Kafka (planned)
• Apache Spark MLlib (planned)
• 4G USB Modem

Final Thoughts

This project connects the physical and digital world by capturing automotive telemetry and enabling intelligent, ML-based diagnostics. It’s an exciting intersection of IoT, machine learning, and real-time streaming—a great example of applying data science beyond traditional dashboards.

If you’re working on similar projects or have suggestions for failure prediction features, feel free to connect or fork the repo!