Research On Algorithm Of 3d Track Detecting Of Objects In 2d Video

The tracking of moving objects is the hotspot of research in computer vision, which largely depends on the fact that the technology has been widely used in intelligent supervising of video, video compression based on content, Virtual Reality, assistant driving and so on. In recent years, many researchers have been dedicated to the field and many algorithms have been proposed to achieve recognition and tracking of objects. For example, the information extracted from contour, texture and color are used, so as the information of the results coming from various algorithms based on Fourier transform or wavelet transform. Meanwhile, other algorithms of parameters estimation such as Kalman Filter and Particle Filter are also introduced into the field.Extracting 3D information from 2D video is also the hotspot in computer vision. 3D contains more interesting information than 2D. The 3D trace of objects could provide more information about objects, which is quite useful in intelligent supervising of video, video compression based on content, Virtual Reality, assistant driving and so on. Extracting 3D information from 2D video is a complex procedure, which should refer to camera imaging model and processing on various noise and error. The most important technology involved in the procedure is the calibration of camera. In recent years, many researchers have been dedicated to the field of calibration of camera.In this paper, the main purpose is to design an algorithm to capture the 3D moving trace of objects in 2D video. An object tracking system which could achieve auto object tracking and get the 3D trace of objects is proposed in this paper. The system consists of two sub modules, namely tracking of moving objects and extracting 3D information. We firstly do a lot of research and analysis on detecting and tracking of objects, camera imaging model, calibrating of camera and coordinates transformation. Then we propose some method to improve the performance of the classical algorithms. Finally, the improved algorithms are united together to fulfill the purpose.As for moving objects tracking, GMM and Particle Filter are mainly used in this paper. Combining the two mathematics methods with objects matching, we achieve an auto moving objects tracking algorithm. First, we use GMM-based background modeling algorithm to detect objects in video. Then the Particle Filter is used to track objects. Finally, the results of objects detecting and tracking are used in objects matching to fulfill the amalgamation and recognition of objects, which could also improve the robustness of the algorithm.The 3D information extracting algorithm is proposed based on research and analysis on camera imaging model and classical algorithm of camera calibration. First, we find the reason which causes the camera calibration algorithm fails. Then a strategy of coordinate modification is proposed according to geometrical information of datum marks in 3D world to avoid the failure of camera calibration. Finally, the 3D coordinates of objects could be calculated with the help of the parameters got by camera calibration.The research in this paper is carried out as an auto object-tracking system which is proposed on PC platform with Visual Studio2005 and OpenCV. The result shows that the system could achieve a good tracking effect, which works better than classical algorithms on small objects and the objects quite like the background. At the same time it could get the precise 3D trace of objects.