Video-based Pose Tracking Of A Car And Motion Simulation

The design and assessment of car model plays a key role in the field of Virtual Manufacturing, and it also determines the sales of the car. Expensive cost of real manufacturing, long cycle period of design hinder the apphcation of real manufacturing. So the virtual manufacturing is becoming the main stream of automobile manufacturing. However, the kinetic model of the car is very complex. In order to assess a car model properly, we must consider much information of the pose and motion of the car.Now, there aresome solutions to simulate the complex motions. One solution use the technology of virtual reality to assess the car model. This method firstly need to model the environment of car, hght of sun and so on. Secondly, it also uses traditional physically-based approach to simulate the kinetic model of car. The two aspects of this method results in the expensive cost of the modeling, what is worse, it can't simulate the complex poses of a car such as launching, turning, acceleration and deceleration. Another solution is to use the augmented reality technology. This technology can realtime merge the virtual model with the video sequence. Also, it can give the observer a real experience of interactivity. However, the camera calibration; moving object tracking, rendering and light capturing which are parts of the augmented reality can't meet the requirements of realtime.To solve the above problems, we propose a new augmented video-based approach. Our method is able to composite the virtual car model into a captured video sequence, which records period of a real moving car. The real car in the original video is then replaced by the rendered virtual car with the same motion and pose. In order to do such a composition seamlessly, we firstly need to restore part of the scene structure, including the track of both the car and the camera as well as the direction of the camera. Secondly, we model the ground and wheel of car. Here, we propose another new indirect-modeling technique to model the ground (in order to generate correct shadow) and the rotation of car wheels, which can be used to further improve the reality of the result video and at the same time, is very easy to be implemented. Thirdly, We restore the lighting condition when capture the real car so that the rendered virtual car model is photo-realistic and more importantly, the composition can be done seamlessly. Compared to the traditional methods, the advantage of our method is obvious and can be concluded as follows:·No physical model is involved and the motion is indicated by the real car implicitly and can be extracted automatically, so our method suffers from no restriction as in physically-based methods.·It is very easy to simulate a lot of kinetic models which are hard to be described through physical model, e.g. the subtle varying of the car pose due to the uneven or bend road, etc.·The result video place the virtual car in the same scene as the real car in original video, which make the assessment to the car model more objective and more accurate.The experimental results show that our augmented video-based method can accurately simulate various pose and motion of wide variety of cars; more over, it can give the observer a true sensory experience, which is very hard to be achieved in traditional methods.