Raw Data Point Cloud Probabilistic Filtering Algorithm

Abstract

Solving the problem of detecting a moving object in a video stream in real time is one of the urgent tasks in computer vision systems. There are various ways, methods and computational algorithms for solving it. One of the promising algorithms for detecting and predicting the position of a moving object is the probabilistic Kalman filter. On the other hand, to detect a moving object and determine the distance to it, the approach of merging lidar and camera sensors is increasingly used. The Kalman filter can be applied as a suitable filtering algorithm capable of handling multiple inputs. This paper proposes a filtering algorithm based on the integration of probabilistic and median data filtering. The advantage of this approach is the replacement of the division operation in computational calculations by the Goldschmidt algorithm. The developed algorithm will reduce the delay time of the algorithm, as well as improve its accuracy. The results obtained can be effectively applied in various computer vision systems.