No abstract

This work aims to enhance a classic method for rigid registration, the iterative closest point (ICP), modifying the closest point search in order to consider approximated information of local geometry combined to the Euclidean distance, originally used. For this, a preprocessing stage is applied, in which the local geometry is encoded in second-order orientation tensors. We define the CTSF, a similarity factor between tensors. Our method uses a strategy of weight variation between the CTSF and the Euclidean distance, in order to establish correspondences. Quantitative tests were made in point clouds with different geometric features, with variable levels of additive noise and outliers and in partial overlapping situations. Results show that the proposed modification increases the convergence probability of the method for higher angles, making the method comparable to state-of-art techniques.

Pairwise rigid registration aims to find the rigid transformation that best registers two surfaces represented by point clouds. This work presents a comparison between seven algorithms, with different strategies to tackle rigid registration tasks. We focus on the frame-to-frame problem, in which the point clouds are extracted from a video sequence with depth information generating partial overlapping 3D data. We use both point clouds and RGB-D video streams in the experimental results. The former is considered under different viewpoints with the addition of a case-study simulating missing data. Since the ground truth rotation is provided, we discuss four different metrics to measure the rotation error in this case. Among the seven considered techniques, the Sparse ICP and Sparse ICP-CTSF outperform the other five ones in the point cloud registration experiments without considering incomplete data. However, the evaluation facing missing data indicates sensitivity for these methods against this problem and favors ICP-CTSF in such situations. In the tests with video sequences, the depth information is segmented in the first step, to get the target region. Next, the registration algorithms are applied and the average root mean squared error, rotation and translation errors are computed. Besides, we analyze the robustness of the algorithms against spatial and temporal sampling rates. We conclude from the experiments using a depth video sequences that ICP-CTSF is the best technique for frame-to-frame registration.

Modern pattern recognition methods are based on convolutional networks since they are able to learn complex patterns that benefit the classification. However, convolutional networks are computationally expensive and require a considerable amount of memory, which limits their deployment on low-power and resource-constrained systems.To handle these problems, recent approaches have proposed pruning strategies that find and remove unimportant neurons (i.e., filters) in these networks. Despite achieving remarkable results, existing pruning approaches are ineffective since the accuracy of the original network is degraded. In this work, we propose a novel approach to efficiently remove filters from convolutional networks. Our approach estimates the filter importance based on its relationship with the class label on a low-dimensional space. This relationship is computed using Partial Least Squares (PLS) and Variable Importance in Projection (VIP). Our method is able to reduce up to 67% of the floating point operations (FLOPs) without penalizing the network accuracy. With a negligible drop in accuracy, we can reduce up to 90% of FLOPs. Additionally, sometimes the method is even able to improve the accuracy compared to original, unpruned, network. We show that employing PLS+VIP as the criterion for detecting the filters to be removed is better than recent feature selection techniques, which have been employed by state-of-the-art pruning methods. Finally, we show that the proposed method achieves the highest FLOPs reduction and the smallest drop in accuracy when compared to stateof-the-art pruning approaches. Codes are available at: https://github.com/arturjordao/PruningNeuralNetworks

In the pairwise rigid registration problem, we need to¯nd a rigid transformation that aligns two point clouds. The classical and most common solution is the Iterative Closest Point (ICP) algorithm. However, the ICP and many of its variants require that the point clouds are already coarsely aligned. We present in this paper a method named Shape-based Weighting Covariance Iterative Closest Point (SWC-ICP) which improves the possibility to correctly align two point clouds, regardless of the initial pose, even when they are only partially overlapped, or in the presence of noise and outliers. It bene¯ts from the local geometry of the points, encoded in second-order orientation tensors, to provide a second correspondences set to the ICP. The cross-covariance matrix computed from this set is combined with the usual cross-covariance matrix, following a heuristic strategy. In order to compare our method with some recent approaches, we present a detailed evaluation protocol to rigid registration. Results show that the SWC-ICP is among the best compared methods, with a better performance in situations of wide angular displacement of noisy point clouds.