Urban environment simulator for train data generation toward CV object recognition

Abstract

Detecting moving objects in an urban environment is a challenging and widely explored problem in computer vision. This task requires huge amounts of data. Their obtaining and labeling is challenging. However the available datasets are not always fit the task. This work proposes a framework for synthesizing the train data based on 3D visualization of an urban environment using Unity 3D. Methods of mathematical statistics and distribution theory were used to build the background models of the framework. The framework, presented in the article, allows to simulate the real urban environment in an adjustable 3D virtual scene. It considers different environmental parameters amd makes possible to simulate the real behavior and physical characteristics of moving objects.