The SCCH develops image-supported analysis systems for the measurement and evaluation of optical patterns as well as spatial-temporal motion data. Here we typically combine various analysis methods: real-time capable image processing on the level of raw data, deep learning for extraction of semantic attributes as intermediate level, and causality and context analysis via statistical and knowledge-based methods.
Examples include inspection systems for surface analysis, objectivization of technical and subjective image quality and, in combination with spatial-temporal motion patterns, the analysis of changes in cells and cell clusters or analysis of tracking data in sports and surveillance systems.
Project Image-Based Knowledge Mining (IKM)
Analysis of game strategies, player positions, travel paths, velocities and acceleration curves in combination with biometric player data such as heart and breathing frequencies is state of the art. One such system is LPM (Local Position Measurement) by abatec group's subsidiary inmotiotec. This system was used, e.g., by the Dutch national football team to prepare for the world championship in 2014 and by Red Bull Salzburg for data analysis. In a precursor project, SCCH developed a framework that was integrated in the LPM system. This framework enables the filtering of position data for objects or persons from video sequences. It builds on highly developed image processing methods that provide high precision and flexibility in object tracing and that were optimized regarding real-time conditions. Thus the resulting data can be evaluated already during training and integrated in the training process.
In the current iKM project, we are enhancing available technologies and know-how in order to cover additional sports with different challenges regarding image processing.