One of the popular topics in machine vision nowadays is the three dimensional scene reconstruction using cameras with different placement. This paper presents a complex method, which is able to reconstruct three dimensional moving objects in the observed area based on video streams of two fixed cameras from any viewpoint between these cameras. It overviews the recent results in the field of machine vision, particularly the related theoretical background and the algorithms in use. This paper comprehensively describes the planning and designing steps of the implemented simulation software used to solve the problem at hand. Using this application one can calibrate the utilized cameras and perform a continuous reconstruction based on the acquired video streams, where the desired viewpoint can be changed with a trackbar on the user interface. In the end it explores and discusses the limitations of the real-time processing and concludes that the implemented solution can also be applied in practice.