Camera Acquisition, Synchronization And Three-dimensional Particle Trajectories

One of the most important function for computer vision is utilizing imaging sensor to acquire the light information of the environment and obtaining physical measurements about the scene through proper analysis. Usually, the image sensor is a camera. In this sense, the camera is serving as an instrument. Recently, using multiple cameras to obtain the 3D structure has becoming a popular subject in the computer vision research community. This thesis proposes new ideas relating to two problems,1) how to synchronize two commercial video cameras and 2) how to obtain the 3D trajectory of each individual in a particle swarm.The contribution of this thesis include:proposing a random binary signal based method to achieve synchronization of sub-frame accuracy among two high frame rate video cameras; proposing two frameworks to obtain the 3D trajectories for particle swarm. In the first framework, the trajectory reconstruction is treated as selecting multiple trajectories through the 3D candidate positions of all time steps. The selection is subject to three constraints. In the second framework, the problem is divided into three sub-problems, each of which is formulated as an Linear Assignment Problem(LAP). Full length trajectories are produced by solving the three LAP subsequently. Both methods can produce accurate trajectories for swarm consisting of hundreds of fruit flies.