Yard management with real-time AI-powered container locator

Logistic sites are often beehives with lots of in- and outbound flows. In such an environment, mistakes are inevitable. And they are costly, for a truck could very well be 1000 kilometres down south before anyone notices it’s carrying the wrong trailer.

Thanks to vision learning and the latest generation of smart cameras, we built Real-time Trailer Location Services to localise trucks anywhere in a yard, day and night. Pioneering a project for H.Essers in Genk, Belgium, we trace vehicles, evaluate movements, identify inefficiencies and optimise (un)loading times with AI that’s 97% accurate.

Figure 1: Logistic sites with lots of in- and outbound flows

“Using satellite imagery of the whole yard and the data of multiple cameras, we create a real-time digital twin of the yards.”

Roshnee Devi Naveen, AI Vision Engineer at Robovision

Yard management is a complex matter—and that’s exactly why we like it. First of all, it’s not a controlled environment. Weather and light conditions are ever-changing, yet fleet managers need to know which truck is where 24/7. To tackle this issue, we collect samples of all possible variations of rain or shine, night and day. Next, we use data augmentation to simulate even more possible conditions, from sleet over a low winter sun to mist. 

Another issue is the similarity of trucks and trailers. Based on spatial and temporal information, we determine which trailer is where at a particular time. Compiling that data with a trailer’s features, we have robust information about its location and can track its every move on the yard, as well as that of hundreds of other, almost identical trailers.

Figure 2: The AI-based locator can track almost identical trailers

Seeing everything, of course, is hard. Because of weather conditions, because of the sheer number of trailers in a yard, but also because seen from ground level, one trailer can hide another. That’s why we set up multiple cameras overhead, creating a 360° view of the whole yard, and train our AI to track roofs, as they are always in sight. Combining all of these features, we can pinpoint a trailer anywhere on the yard with a 1-metre accuracy—a result that by far beats GPS coordinates.

Figure 3: Multiple cameras were set up overhead to create a 360° view of the whole yard

“The results prove that this technology is scalable. Vision learning will disrupt the supply chain as we know it.”

Matti Desmet, Product & Engineering Director at Peripass

Building a detection system with such a low margin for error is always teamwork. We rely on the specific experience of our partners in the particularities of yard management on the one hand, and smart cameras on the other. At H.Essers, AeyeQ mounted their Augur Vehicle, a powerful Edge camera that collects and processes visual data, which we use as a basis for deep learning.

Next, on-site operators, dispatchers and logistics managers use their knowledge to label the data through our tool, using bounding polygons. This, in turn, connects to Peripass’ yard management application and creates a real-time twin of the yard.

Figure 4: The Robovision Platform was used to label data and train a deep-learning model

Together, we’ve built a tool that acts as an extra pair of eyes in the sky and detects incidents as they occur, from a truck docking in the wrong place to a truck picking up the wrong trailer. We can even track the number of rented trailers parked on-site to avoid rent fees from accumulating. On a yard the size of 25 football fields, there’s no chance humans could match our feat.

Figure 5: Final product – a real-time AI-based tracking system for smart yard management