ANALYSIS OF IMAGES OBTAINED BY LASER SCANNING OF SPACE IN LOW VISIBILITY CONDITIONS USING VORONOI DIAGRAMS

Abstract

Introduction. Processing of space scanning data by lidar has features due to the principle of operation of the device. Objects located closer shade those located behind them. This also happens in the case of obstacles in the form of smoke and other similar phenomena. Points recorded as a result of reflection from smoke create noise interference and worsen the clarity of images in the background. Removing them due to narrowing the distance ranges can lead to the loss of images of real objects located in the foreground. The use of traditional clustering methods has both positive aspects and disadvantages, due to the need to choose a method and select input parameters. Therefore, it is relevant to use other methods of analysis and image improvement for identifying real objects and removing noise in smoke conditions, in particular the Voronoi diagram method. 
Purpose. The purpose of the work is to improve the analysis of images obtained by lidar using the method of determining the local density of points, which is based on the use of Voronoi diagrams. 
Methods. Methods for analyzing discrete 3D images obtained with the Intel RealSense L515 lidar in normal conditions and in low visibility (smoke), using Voronoi diagrams. The method is implemented in Python. Interval statistical distributions are used for analysis. 
Results. Using Voronoi diagrams, it is possible to determine the local density in the Voronoi cell of each point of the data set. To avoid image clutter and better distinguish local density during visualization, it is advisable to display this indicator by the color of the points on a logarithmic scale. The method is implemented in Python for two sets of points: in normal conditions and in smoke. In the sets of points ranked by local density, those with the largest and smallest of these indicators are discarded. It was found that noise was removed, in particular caused by smoke, after discarding points with the lowest local density and images of real objects were improved by discarding points with the highest local density. Since a real small object in 3D space may be lost after a such procedure, there is a need to choose the optimal proportion of points to discard. Comparison of interval distributions by local density for both data sets revealed an increase in the number of points on the interval close to the maximum density and a decrease on the interval preceding it. 
Conclusions. The Voronoi diagram method makes it possible to detect cluster points corresponding to real objects and reduce noise by discarding points with the lowest density. Small real objects in a smoky environment are more difficult to detect due to the fact that they are reflected by a smaller number of points and lower local densities for boundary points, which requires the selection of the proportion to discard.