Adaptation of the machine vision system to environmental conditions

Abstract

The procedure for the formation of the frame processing algorithm training model in a machine vision system designed to solve the visual odometry issue has been considered. The main difficulties arising in matching the stereopair frames have been described. The scheme and the set of frame processing features have been proposed. The work of the teacher algorithm based on the neural network has been considered. The options of the algorithm adaptation to an applied task have been analyzed. The options to implement the system and the results of computational and field experiments have been presented.