Application Research Of 3D Ladar Image In Railway Foreign Body Invasion Detection

Intrusion detection of foreign bodies in railway has been a hot research topic for many years.Most of the invasion detection methods adopt visual methods,because the vision is greatly affected by the changes of external environment,such as illumination,weather and so on.In this paper,3D laser radar railway foreign body invasion detection system is proposed.The system is designed and realized,which can restore the 3D scene and realize the target detection.The 3D laser radar railway foreign body intrusion detection system mainly includes the design of 3D radar,3D point cloud data acquisition,point cloud data preprocessing and target detection.The main work of this paper includes:1)the realization of cloud head system and the design of controller hardware circuit.The 3D radar system is realized by using the GL-1025 two-dimensional radar and the cloud head system.The interface of upper computer is designed to realize the communication between GL-1025 radar and cloud head system,to determine the scanning sequence of the two systems,to transform the format of the collected 3D point cloud data and to model the 3D model.4)preprocessing the 3D point cloud data,because the point cloud data is too large,it adopts dimensionality reduction processing,and determines the detection area and removes the outlier.Point cloud processing adopts two schemes:grayscale image processing scheme and point cloud clustering scheme.1)grayscale image processing scheme:based on the height feature of the target,the method of maximum Inter-class Variance(OTSU)is used to find the suitable threshold to separate the target from the background.Experimental results show that the algorithm can isolate some targets,but it is easy to lose small targets.There are many methods of point cloud clustering,which are compared by K-means clustering and DBSCAN(Density-Based Spatial Clustering of Applications with Noise)clustering.Experiments show that K-means clustering can not separate the target from the background DBS CAN density clustering algorithm has obvious advantages,can be automatically classified,and has a high detection accuracy in multiple scenes.