Simulation of pedestrian dynamics based withemantic trajectory segmentation

Abstract

The article analyzes the existing methods of information processing necessary for the functioning of the system of intelligent control over unregulated pedestrian crossings based on aggregation and data processing by means of IOT. The state space model of the switching Kalman filter is considered, the development of mathematical software for the analysis and processing of information based on the results of intelligent control over unregulated pedestrian crossings, in particular with semantic segmentation of trajectories using agent-based models, is carried out. An MDA (Markov Decision Process) state space model is presented, a Hidden Markov Model (HMM) which has discrete hidden variables. The developments for the development of the following subsystems are presented: activity detector subsystem. Receives video frames as input, supports the static object model (background model) and returns the hotspot mask for the current frame; subsystems for detecting and tracking objects (pedestrians and cars). Based on the video frame and hotspot mask, it detects and accompanies objects of a given class, returning their coordinates; trajectory analysis subsystem. Analyzing the history of movement of pedestrians and cars, returns the facts of traffic violations.