Channel impulse response based vehicle analysis in tunnels

Indoor localization and positioning is of vital importance in numerous applications. In particular, in the case of emergency events, locating and tracking of the victims, objects and rescue personnel in harsh indoor environments is still challenging. In this paper, two different approaches for the obstruction detection inside the road tunnel are analysed. Both methods are based on the analysing channel impulse responses (CIRs). The first parametric approach tests the use of root mean squared signal delay spread to recognize the object in an empty tunnel. Because the recognition of additional objects in already occupied tunnel is unreliable, more complex machine learning approach is also tested. The convolutional neural network (CNN) classification model for the LoS/NLoS channel detection is able to detect the object in an empty tunnel with the accuracy of more than 90%, whereas the presented multiple objects scenarios can be successfully resolved in more than 80%.