One of the main problems in the elderly population and for people with functional disabilities is falling when they are not supervised. Therefore, there is a need for monitoring systems with fall detection functionality. Mobile robots are a good solution for keeping the person in sight when compared to static-view sensors. Mobile-patrol robots can be used for a group of people and systems are less intrusive than ones based on mobile robots. In this paper, we propose a novel vision-based solution for fall detection based on a mobile-patrol robot that can correct its position in case of doubt. The overall approach can be formulated as an end-to-end solution based on two stages: person detection and fall classification. Deep learning-based computer vision is used for person detection and fall classification is done by using a learning-based Support Vector Machine (SVM) classifier. This approach mainly fulfills the following design requirements—simple to apply, adaptable, high performance, independent of person size, clothes, or the environment, low cost and real-time computing. Important to highlight is the ability to distinguish between a simple resting position and a real fall scene. One of the main contributions of this paper is the input feature vector to the SVM-based classifier. We evaluated the robustness of the approach using a realistic public dataset proposed in this paper called the Fallen Person Dataset (FPDS), with 2062 images and 1072 falls. The results obtained from different experiments indicate that the system has a high success rate in fall classification (precision of 100% and recall of 99.74%). Training the algorithm using our Fallen Person Dataset (FPDS) and testing it with other datasets showed that the algorithm is independent of the camera setup.

A very fast filter bank optimization technique is proposed for synthesizing efficient nearly perfect-reconstruction cosine-modulated transmultiplexer systems for multicarrier modulation. The most important problems in communications systems are the existence of both the ISI and inter-carrier interference. In order to properly minimize both of these, a windowing technique is used for generating the prototype filter for the cosine-modulated synthesis and analysis filter banks in such a way that a proper weighted sum of the above-mentioned interferences in the sub-channels is minimized. For this purpose, the parameters of the generalized Blackman window proposed by the authors of this paper and the cutoff frequency of the ideal filter are optimized subject to the condition that the 3-dB cutoff frequency of the prototype filter occurs at /spl omega/=/spl pi//(2M) with M being the number of channels. Numerical simulations show that the proposed technique leads to an increase in the signal-to-interference ratio compared to the original Blackman window giving the best results, according to our experience, among the well-known windows.

In this work, we introduce an intelligent video sensor for the problem of Action Proposals (AP). AP consists of localizing temporal segments in untrimmed videos that are likely to contain actions. Solving this problem can accelerate several video action understanding tasks, such as detection, retrieval, or indexing. All previous AP approaches are supervised and offline, i.e., they need both the temporal annotations of the datasets during training and access to the whole video to effectively cast the proposals. We propose here a new approach which, unlike the rest of the state-of-the-art models, is unsupervised. This implies that we do not allow it to see any labeled data during learning nor to work with any pre-trained feature on the used dataset. Moreover, our approach also operates in an online manner, which can be beneficial for many real-world applications where the video has to be processed as soon as it arrives at the sensor, e.g., robotics or video monitoring. The core of our method is based on a Support Vector Classifier (SVC) module which produces candidate segments for AP by distinguishing between sets of contiguous video frames. We further propose a mechanism to refine and filter those candidate segments. This filter optimizes a learning-to-rank formulation over the dynamics of the segments. An extensive experimental evaluation is conducted on Thumos’14 and ActivityNet datasets, and, to the best of our knowledge, this work supposes the first unsupervised approach on these main AP benchmarks. Finally, we also provide a thorough comparison to the current state-of-the-art supervised AP approaches. We achieve 41% and 59% of the performance of the best-supervised model on ActivityNet and Thumos’14, respectively, confirming our unsupervised solution as a correct option to tackle the AP problem. The code to reproduce all our results will be publicly released upon acceptance of the paper.

The objective of this work is the proposal of a new navigation algorithm and its integration in a platform that is already designed and built, improving the functionality of the robot in order to patrol complex indoor environments. This patrol contains various features related to the navigation and the localization of the platform using a particle filter that allows the robot to move autonomously through the environment with the data obtained from an RGB-D camera and LIDAR. The navigation algorithm is adapted dynamically in real-time using the well-known CNN real-time object detector You Only Look Once (YOLOv3), which we have retrained with our own database. The platform detects standing and fallen people. Additionally, it registers people using a specific face recognition convolutional neural network. All these functionalities are controlled and centralized in a friendly user interface that appears on the robot’s touch screen and a voice service model is also used.