Action Classification for Partially Occluded Silhouettes by Means of Shape and Action Descriptors

This paper presents an action recognition approach based on shape and action descriptors that is aimed at the classification of physical exercises under partial occlusion. Regular physical activity in adults can be seen as a form of non-communicable diseases prevention, and may be aided by digital solutions that encourages individuals to increase their activity level. The application scenario includes workouts in front of the camera, where either the lower or upper part of the camera’s field of view is occluded. The proposed approach uses various features extracted from sequences of binary silhouettes, namely centroid trajectory, shape descriptors based on the Minimum Bounding Rectangle, action representation based on the Fourier transform and leave-one-out cross-validation for classification. Several experiments combining various parameters and shape features are performed. Despite the presence of occlusion, it was possible to obtain about 90% accuracy for several action classes, with the use of elongation values observed over time and centroid trajectory.

Representation and Reasoning of Three-Dimensional Spatial Relationships Based on R5DOS-Intersection Model Representation and Reasoning Based on R5DOS Model

This paper aims to disclose the compound topological and directional relationships of three simple regions in the three-dimensional (3D) space. For this purpose, the directional model and the 8-intersection model were coupled into an R5DOS-intersection model and used to represent three simple regions in the 3D space. The matrices represented by the model were found to be complete and mutually exclusive. Then, a self-designed algorithm was adopted to solve the model, yielding 11,038 achievable topological and directional relationships. Compared with the minimum bounding rectangle (MBR) model, the proposed model boasts strong expressive power. Finally, our model was applied to derive the topological and directional relationships between simple regions A and C from the known relationships between simple regions A and B and those between B and C. Based on the results, a compound relationship reasoning table was established for A and C. The research results shed new light on the representation and reasoning of 3D spatial relationships.

Quadrant-Based Minimum Bounding Rectangle-Tree Indexing Method for Similarity Queries over Big Spatial Data in HBase

With the rapid development of mobile devices and sensors, effective searching methods for big spatial data have recently received a significant amount of attention. Owing to their large size, many applications typically store recently generated spatial data in NoSQL databases such as HBase. As the index of HBase only supports a one-dimensional row keys, the spatial data is commonly enumerated using linearization techniques. However, the linearization techniques cannot completely guarantee the spatial proximity of data. Therefore, several studies have attempted to reduce false positives in spatial query processing by implementing a multi-dimensional indexing layer. In this paper, we propose a hierarchical indexing structure called a quadrant-based minimum bounding rectangle (QbMBR) tree for effective spatial query processing in HBase. In our method, spatial objects are grouped more precisely by using QbMBR and are indexed based on QbMBR. The QbMBR tree not only provides more selective query processing, but also reduces the storage space required for indexing. Based on the QbMBR tree index, two query-processing algorithms for range query and kNN query are also proposed in this paper. The algorithms significantly reduce query execution times by prefetching the necessary index nodes into memory while traversing the QbMBR tree. Experimental analysis demonstrates that our method significantly outperforms existing methods.

DETECTION OF HARBOURS FROM HIGH RESOLUTION REMOTE SENSING IMAGERY VIA SALIENCY ANALYSIS AND FEATURE LEARNING

Harbours are very important objects in civil and military fields. To detect them from high resolution remote sensing imagery is important in various fields and also a challenging task. Traditional methods of detecting harbours mainly focus on the segmentation of water and land and the manual selection of knowledge. They do not make enough use of other features of remote sensing imagery and often fail to describe the harbours completely. In order to improve the detection, a new method is proposed. First, the image is transformed to Hue, Saturation, Value (HSV) colour space and saliency analysis is processed via the generation and enhancement of the co-occurrence histogram to help detect and locate the regions of interest (ROIs) that is salient and may be parts of the harbour. Next, SIFT features are extracted and feature learning is processed to help represent the ROIs. Then, by using classified feature of the harbour, a classifier is trained and used to check the ROIs to find whether they belong to the harbour. Finally, if the ROIs belong to the harbour, a minimum bounding rectangle is formed to include all the harbour ROIs and detect and locate the harbour. The experiment on high resolution remote sensing imagery shows that the proposed method performs better than other methods in precision of classifying ROIs and accuracy of completely detecting and locating harbours.