Metric reconstruction of multiple rigid objects

[摘要] Engineers struggle to replicate the capabilities of the sophisticated human visualsystem. This thesis sets out to recover motion and 3D structure of multiple rigidobjects up to a similarity. The motion of these objects are either recorded in asingle video sequence, or images of the objects are recorded on multiple, di erentcameras. We assume a perspective camera model with optional provision forcalibration information. The Structure from Motion (SfM) problem is addressedfrom a matrix factorization point of view. This leads to a reconstruction correctup to a projectivity of little use in itself. Using techniques from camera autocalibrationthe projectivity is upgraded to a similarity. This reconstructionis also applied to multiple objects through motion segmentation. The SfMsystem developed in this thesis is a batch-processing algorithm, requiring fewframes for a solution and readily accepts images from very di erent viewpoints.Since a solution can be obtained with just a few frames, it can be used toinitialize sequential methods with slower convergence rates, such as the Kalman lter. The SfM system is critically evaluated against an extensive set of motionsequences.

[发布机构] Stellenbosch University