Modulai developed an object detection model for Fuelmatics' fully automatic vehicle refueling system. The model detects and locates fuel lids on cars in real time, and because it is optimized to run on a CPU, the fueling rig operates autonomously without internet connectivity or GPU hardware. Samma fakta, men "CPU-optimized" får sin betyde
Challenge
Fuelmatics is developing a fully automatic vehicle refueling system. In some parts of the world, drivers do not refuel their own cars, for cultural, practical, or legislative reasons. Having staff do it is expensive and inefficient, and in extreme heat it is outright dangerous. Automating the process means the device has to identify and locate the fuel lid quickly and reliably.
Solution
Modulai collected and annotated images of fuel lids across a range of car models, then evaluated several object detection candidates on both inference speed and prediction accuracy. The chosen model was integrated into the fueling rig's central control system as its main detection mechanism, first verifying that the car is correctly positioned, then guiding the fueling arm to the fuel lid in real time. The whole system runs on-device, without internet connectivity.
Tools
The models used were trained using the TensorFlow Object Detection API. In this case, we resorted to using a pre-trained lightweight object detection model that we fine-tuned. The model was CPU-optimized to be run efficiently on GPU as well as on CPU.
Value created
By keeping the driver inside the car and delegating the labor to a robotic arm, it eliminates human error, prevents fuel spills, and significantly cuts down on wait times. For station operators, that speed translates into more vehicles served per hour and a service they can offer at a premium. Modulai's detection model is a core part of what makes this possible, locating the fuel lid quickly and reliably on standard hardware, so the rig can operate on its own without a network connection.
Learn more
