Pipeline surveillance and perimeter security using laser backscatter radiation in glass fibers are current fields of research. The outstanding opportunity - opposing to e.g. methods applying interference patterns - of using this method is the very accurate temporal and spatial localization of events on the glass fiber. However, the vast amount of data, collected inside these setups and its subsequent analysis presents a problem for today's systems. Further, as the data are highly dependent on the fiber and material wherein it is located, they are not easily interpreted.
The backscattered radiation can be interpreted like an audio record. So in our work we utilize artificial intelligence methods to classify this "audio" stream to assign meaningful labels to specific recorded events. Depending on the method in use or using ensemble methods, this classification process is quite time consuming. To simplify this task, first we identify the regions of interest (ROI) that deserve further investigation. To facilitate this, we apply a preprocessing algorithm on the raw data.
Using the classification labels we further define objects and enrich them with additional details, such as their movement along the cable. Depending on the sampling rate, even groups of objects with small gaps between them - e.g. a convoy of cars or people walking behind each other - become distinguishable. Using sonic velocity information, we also are able to determine the rectangular
distance of the observed object to the fiber. This information is then used to detect approaching objects before they reach a certain perimeter limit.