This project was a machine-vision proof of concept delivered to a large multinational commodity distribution company in Western Canada. The company had recently implemented an automated railcar unloading system, and this POC aimed to demonstrate how a machine-learning–based object detection solution could augment the existing system by accurately identifying, localising and determining the orientation of capstan sockets on incoming train cars.
Traditional vision-based automation struggled due to the variability in railcar geometry, capstan designs, lighting conditions, debris and other environmental factors. The goal of the POC was to show that a well-trained machine learning model could outperform these legacy methods – delivering high accuracy, low latency and a clean, structured data stream suitable for real-time integration into the existing control system.
The existing system had acquisition times exceeding 1 second and a success rate below 80%. For the POC, still images and video of multiple capstan types were captured, labelled and used to train a custom model. The resulting model successfully detected capstans on 100% of the validation set with confidence levels consistently above 85% and was able to infer both location and rotation. When evaluated on live video, it achieved acquisition times under 25 ms with zero false positives on surrounding train structures.
