The design of a robust 3D Reconstruction system for video sequences in non controlled environments
Document typeMaster thesis
Rights accessOpen Access
Along this thesis, a novel and robust approach for obtaining 3D models from video sequences captured with hand-held cameras is adressed. This work defines a fully automatic pipeline that is able to deal with diferent types of sequences and acquiring devices. The designed and implemented system follows a divide and conquer approach. An smart frame decimation process reduces the temporal redundancy of the input video sequence and selects the best conditioned frames for the reconstruction step. Next, the video is split into overlapped clips with a fixed and small number of Key-frames. This allows to parallelize the Structure and Motion process which translates into a dramatic reduction in the computational complexity. The short length of the clips allows an intensive search for the best solution at each step of the reconstruction, which improves the overall system performance. The process of feature tracking is embedded within the reconstruction loop for each clip as a difference with other approaches. The last contribution of this thesis is a final registration step that merges all the processed clips to the same coordinate frame. This last step consists on a set of linear algorithms that combine information of the structure (3D points) and motion (cameras) shared by partial reconstructions of the same static scene to more accurately estimate their registration to the same coordinate system. The performance for the presented algorithm as well as for the global system is demonstrated in experiments with real data.