Towards Automated Floorplan Generation
This pipeline is used for generating 2D floorplans using inexpensive RGB-D sensors such as the Kinect. Using ORB-SLAM, the camera trajectories and the 3D point clouds of the enclosures are obtained. The huge volumetric representations are abstracted to obtain rectilinear representations and then subsequently used for generating 2D floorplans.
2. Robust Relative Motion Averaging
Obtaining global camera motions from the pairwise motions of images/pointclouds is a key component in global 3D-reconstruction pipelines. My work focused on developing accurate and scalable algorithms for doing Rotation Averaging used to estimate the global camera rotations and developing theoretical foundations under which the algorithms can converge to a good local minima. An efficient algorithm for estimating the global translations given the translation directions is also developed (popularly known as Translation Averaging). The results indicate that these algorithms outperform the state-of-the-art methods in accuracy and efficiency.
3. Adaptive Annealing for Robust Geometric Estimation
Graduated Non-Convexity (GNC), a popular technique used in minimization of statistical loss functions for robust estimation of parameters under noise and outliers, attempts to avoid local minima by annealing the scale parameter of the loss functions. Often, this annealing is carried out in a fixed manner, resulting in a poor speed-accuracy trade-off and unreliable convergence to the global minimum. In this paper, we propose a principled approach for adaptively annealing the scale for GNC by tracking the positive-definiteness (i.e. local convexity) of the Hessian of the cost function. We illustrate our approach using the classic problem of registering 3D correspondences in the presence of noise and outliers. Our approach is more accurate and efficient and converges to the global solution more reliably than the state-of-the-art methods. More details can be found in the project page.