Virtualmech is developing an on-budget system that really is affordable for railway operators & maintainers to allow infrastructure management measuring from an operational vehicle. The defect detection must be robust and fast, so no important infrastructure damage is overlooked. The robustness and efficient acquisition and treatment of the images while keeping the costs controlled is our main unique selling point that cannot be reached by any of our competitors.
This can only be made with a good expertise in both computer vision and machine learning systems. Digital image processing will allow the detection of the main features in each frame that can then be evaluated by the neural networks. In addition, a good training of the neural network is essential, as the quantity of information that is going to be generated is really huge (RGB 2D pictures @ 100 Hz) and the only way to drastically reduce this data size is by filtering pictures as they are taken.