In response to a growing number of vehicle thefts affecting truck and van owners, Ford approached our team to develop a real-time threat detection system. We developed a fully custom, sensor-fusion AI, leveraging sound and motion to detect vehicle break-ins before they happen. It was designed with severe power constraints to continuously run at the edge, requiring us to innovate across all stages of the model development cycle.

Data collection and QA

We conducted a robust training data collection campaign with specialists simulating vehicle break-ins to capture, annotate, and test thousands of audio and motion samples.

R&D, model architecture

We began by studying how vehicle break-ins typically occur, especially in trucks and vans. This helped us identify the most effective sensors and determine how to apply machine learning in a practical way. Based on that research, we built a system designed to detect clear signs of a break-in while avoiding false alarms caused by normal environmental activity.

We applied advanced sensor fusion and AI techniques to create high-performing binary and multi-class classifiers while significantly reducing false positives. Packaged into a state-of-the-art threat detection algorithm.

Embedded Engineering

Our embedded engineering team collaborated with top AI chip manufacturers to optimize models for low-power edge environments, enabling seamless integration into Ford’s hardware development cycle.