ALGORITHM FOR INCREASING THE SPATIAL DENSITY OF A LIDAR POINT CLOUD FOR SOLVING PROBLEMS OF AUTONOMOUS DRIVING
Annotation
Several algorithms are proposed to solve the problem of increasing the density of points of lidar clouds, including the algorithms based on neural networks. High density lidar clouds can improve the accuracy of scene recognition and localization algorithms. The algorithms quality is assessed by the following metrics: the number of false points, the average error, the root-mean-square error. In contrast to existing ones, the proposed algorithms are aimed at determining not only additional points positions, but also their reflectivity coefficients. Presented results of experiments demonstrate that the neural network algorithm without the use of normalization and sigmoidal weighting of the loss function has the smallest error. The least number of false points is provided by the neural network algorithm with an added normalization.
Keywords
Постоянный URL
Articles in current issue
- DIGITAL TWINS IN INDUSTRY: DEVELOPMENT HISTORY, CLASSIFICATION, TECHNOLOGIES, USE CASES
- APPLYING MACHINE LEARNING METHODS TO PREDICT OR REPLACE MISSING LOGGING DATA
- USING SUBBAND COSINE TRANSFORM MATRICES TO IMPROVE THE VISUAL QUALITY OF OBJECTS CONTOURS IN AEROSPACE IMAGES
- CONSTRUCTION, MODELS, AND SIGNALS PROCESSING OF MULTIFUNCTIONAL RECEIVER IN AIRCRAFT AEROMETRIC SYSTEMS
- NEW TOOLS FOR MONITORING THE DYNAMIC ERRORS OF MAGNETIC COMPASSES
- APPLICATION OF FLUORESCENCE SPECTROSCOPY FOR in vitro ANALYSIS OF FILLING MATERIALS AND HARD TOOTH TISSUES
- CHARACTERISTICS AND PARAMETERS OF THE TECHNICAL CONDITION OF OVERHEAD POWER LINES
- HARDWARE AND SOFTWARE COMPLEX FOR SORTING AND CLASSIFICATION OF MINERAL RAW MATERIALS BY VISUAL INDICATORS