Real time light-independent slope-failure monitoring using LiDAR and 2D-3D semantic segmentation

Slope stability mapping plays a significant role in risk management of mountainous area traffic, grand water dam, open cast mine and near slope residence, etc. Yet, real time slope-failure monitoring is one of the most challenging task. Particularly, most prevalent real-time slope monitoring methods are light dependent, scanning slope failure in dark or weak light environment has not been fully investigated so far. In this work, we proposed a heterogeneous sensor data fusion model using LiDAR and visible optic camera. Our proposal combines 2D-3D semantic segmentation and target tracking algorithm to enable 24-hours real-time monitoring and evaluation of typical slope-failures without concern of illuminant factors. Experiments demonstrated the effectiveness of our proposal with quality performance in terms of both efficiency and accuracy.