Deep learning based virtual point tracking for real-time target-less dynamic displacement measurement in railway applications

In the application of computer-vision based displacement measurement, an optical target is usually required to prove reference. case that cannot be attached measuring objective, edge detection, feature matching and template are most common approaches in target-less photogrammetry. However, their performance significantly relies on parameter settings. This becomes problematic dynamic scenes where complicated background texture exists varies over time. To tackle this issue, we propose virtual point tracking for real-time incorporating deep learning techniques domain knowledge. Our approach consists three steps: 1) automatic calibration detection region interest; 2) each video frame using convolutional neural network; 3) domain-knowledge rule engine adjacent frames. The proposed can executed computer a manner (i.e. 30 frames per second). We demonstrate our railway application, lateral wheel rail measured during operation. also implement algorithm line as baseline comparison. numerical experiments have been performed evaluate latency harsh environment with noisy varying backgrounds.