Research On Panoramic 3D Object Tracking Method Based On Spherical Camera Model

Panoramic 3D object tracking utilizes panoramic vision technology to capture omnidirectional image, having advantages of wide field of view and information integrity based on object tracking and stereo vision of panoramic vision. In addition, with the purpose of acquiring 3D information while object tracking, stereo vision technology based on panoramic image is studied. Panoramic 3D object tracking is capable of improving deficiencies and performance of the traditional 2D object tracking and helping with growing requirements. In this thesis, the relevant theories and methods of panoramic 3D object tracking were thoroughly studied, and a binocular spherical camera model was constructed. Besides, object tracking and stereo vision that based on spherical panoramic image were studied. Based on the above studies, panoramic 3D object tracking was accomplished.Firstly, the mapping relationship between the pinhole camera model and the spherical camera model was introduced, and the spherical camera model was constrcuted based on multi-camera system. Futhermore, a vertical synclastic structure was designed as the hardware base of the integral algorithm.Secondly, the spherical panoramic image stereo matching algorithm was studied in this thesis. Thereby, a spherical SURF (Speeded up robust features) algorithm was proposed by combining the standard SURF algorithm with the characteristics of spherical panoramic image. Besides, some concepts, such as the spherical epipolar geometry, the spherical essential were introduced. Based on these concepts and RANSAC (Random sample consensus) algorithm, a mismatch removing methods was accomplished.Thirdly, a fast accurate object tracking algorithm was studied. Compressed sensing theory was introduced and used to extracting features in this thesis. Then, MIL (Multiple instance learning) Boosting algorithm which was used to combine the classifiers was expounded, and the algorithm was improved by using temporal and spatial constraint. Finally, the object tracking which is based on detection was achieved.At last, the method of 3D information compute based on spherical panoramic image was introduced. Then, depth of field compute of binocular stereo vision and object tracking, both of which were based on spherical panoramic image were realized. Subsequently, the integrated framework and the detailed process of the proposed algorithm were designed. The experiment results show that the algorithm proposed in this thesis can track objects in a wide field of view and acquire objects’ 3D information simultaneously.