In this paper, we reconstruct 3D objects with a heterogeneous sensor network of Range Imaging(RIM) sensors and high-res camcorders. With this setup, we first carry out simple but effective depth calibration for the RIM cameras. We then combine the camcorder silhouette cues and RIM camera depth information, for the reconstruction. Our main contribution is the proposal of a sensor fusion framework so that the computation is general, simple and scalable. Although we only discuss the camcorders and RIM cameras in this paper, the proposed framework can be applied to any type of vision sensors. It uses a space occupancy grid as a probabilistic 3D representation of scene contents. After defining sensing models for each type of sensors, the reconstruction is simply a Bayesian inference problem, and can be solved robustly. The experiments show that the recover full 3D closed shapes substantially improved the quality of the noisy RIM sensor measurement.

We are interested in fitting a surface model such as a tensor-product spline to range image data. This is commonly done by finding control points which minimize a compound cost including the goodness of fit and a regularizer, balanced by a regularization parameter. Many approaches choose this parameter as the minimizer of, for example, the cross-validation score or the L-curve criterion. Most of these criteria are expensive to compute and difficult to minimize. We propose a novel criterion, the L-tangent norm, which overcomes these drawbacks. It gives sensible results with a much lower computational cost. This new criterion has been successfully tested with synthetic and real range image data.

Many computer vision tasks require interest point detection and description, such as real-time visual navigation. We present a GPU implementation of the recently proposed Speeded-Up Robust Feature extractor, currently the state of the art for this task. Robust feature descriptors can give vast improvements in the quality and speed of subsequent steps, but require intensive computation up front that is well-suited to inexpensive graphics hardware. We describe the algorithm's translation to the GPU in detail, with several novel optimizations, including a new method of computing multi-dimensional parallel prefix sums. It operates at over 30 Hz at HD resolutions with thousands of features, and in excess of 70 Hz at SD resolutions.

The volume of raw range image data that is required to represent just a single scene can be extensive; hence direct interpretation of range images can incur a very high computational cost. Range image segmentation and feature extraction have been identified as mechanisms to produce more compact scene representations and hence enable less costly scene interpretation for applications such as object recognition and robot navigation. We present a new approach to multi-scale edge and planar surface detection in range images that can be used directly with any range data, regardless of whether the data have regular or irregular spatial distribution. The approach is evaluated with respect to accuracy of both edge location and planar surface representation. The contribution of our approach is that only a single application of our operator is necessary to both extract object edges and determine representations of the corresponding planar object surfaces.