STAC: Spatial-Temporal Attention on Compensation Information for Activity Recognition in FPV.

Egocentric activity recognition in first-person video (FPV) requires fine-grained matching of the camera wearer's action and the objects being operated. The traditional method used for third-person action recognition does not suffice because of (1) the background ego-noise introduced by the unstructured movement of the wearable devices caused by body movement; (2) the small-sized and fine-grained objects with single scale in FPV. Size compensation is performed to augment the data. It generates a multi-scale set of regions, including multi-size objects, leading to superior performance. We compensate for the optical flow to eliminate the camera noise in motion. We developed a novel two-stream convolutional neural network-recurrent attention neural network (CNN-RAN) architecture: spatial temporal attention on compensation information (STAC), able to generate generic descriptors under weak supervision and focus on the locations of activated objects and the capture of effective motion. We encode the RGB features using a spatial location-aware attention mechanism to guide the representation of visual features. Similar location-aware channel attention is applied to the temporal stream in the form of stacked optical flow to implicitly select the relevant frames and pay attention to where the action occurs. The two streams are complementary since one is object-centric and the other focuses on the motion. We conducted extensive ablation analysis to validate the complementarity and effectiveness of our STAC model qualitatively and quantitatively. It achieved state-of-the-art performance on two egocentric datasets.

Hierarchical Optimization of 3D Point Cloud Registration.

Rigid registration of 3D point clouds is the key technology in robotics and computer vision. Most commonly, the iterative closest point (ICP) and its variants are employed for this task. These methods assume that the closest point is the corresponding point and lead to sensitivity to the outlier and initial pose, while they have poor computational efficiency due to the closest point computation. Most implementations of the ICP algorithm attempt to deal with this issue by modifying correspondence or adding coarse registration. However, this leads to sacrificing the accuracy rate or adding the algorithm complexity. This paper proposes a hierarchical optimization approach that includes improved voxel filter and Multi-Scale Voxelized Generalized-ICP (MVGICP) for 3D point cloud registration. By combining traditional voxel sampling with point density, the outlier filtering and downsample are successfully realized. Through multi-scale iteration and avoiding closest point computation, MVGICP solves the local minimum problem and optimizes the operation efficiency. The experimental results demonstrate that the proposed algorithm is superior to the current algorithms in terms of outlier filtering and registration performance.