Filling the Gap between Research and Market: Portable Architecture for an Intelligent Autonomous Wheelchair.

Under the umbrella of assistive technologies research, a lot of different platforms have appeared since the 1980s, trying to improve the independence of people with severe mobility problems. Those works followed the same path coming from the field of robotics trying to reach users' needs. Nevertheless, those approaches rarely arrived on the market, due to their specificity and price. This paper presents a new prototype of an intelligent wheelchair (IW) that tries to fill the gap between research labs and market. In order to achieve such a goal, the proposed solution balances the criteria of performance and cost by using low-cost hardware and open software standards in mobile robots combined together within a modular architecture, which can be easily adapted to different profiles of a wide range of potential users. The basic building block consists of a mechanical chassis with two electric motors and a low-level electronic control system; driven by a joystick, this platform behaves similar to a standard electrical wheelchair. However, the underlying structure of the system includes several independent but connected nodes that form a distributed and scalable architecture that allows its adaptability, by adding new modules, to tackle autonomous navigation. The communication among the system nodes is based on the controller area network (CAN) specification, an extended standard in industrial fields that have a wide range of low-cost devices and tools. The system was tested and evaluated in indoor environments and by final users in order to ensure its usability, robustness, and reliability; it also demonstrated its functionality when navigating through buildings, corridors, and offices. The portability of the solution proposed is also shown by presenting the results on two different platforms: one for kids and another one for adults, based on different commercial mechanical platforms.

Semi-Supervised Anomaly Detection in Video-Surveillance Scenes in the Wild.

Surveillance cameras are being installed in many primary daily living places to maintain public safety. In this video-surveillance context, anomalies occur only for a very short time, and very occasionally. Hence, manual monitoring of such anomalies may be exhaustive and monotonous, resulting in a decrease in reliability and speed in emergency situations due to monitor tiredness. Within this framework, the importance of automatic detection of anomalies is clear, and, therefore, an important amount of research works have been made lately in this topic. According to these earlier studies, supervised approaches perform better than unsupervised ones. However, supervised approaches demand manual annotation, making dependent the system reliability of the different situations used in the training (something difficult to set in anomaly context). In this work, it is proposed an approach for anomaly detection in video-surveillance scenes based on a weakly supervised learning algorithm. Spatio-temporal features are extracted from each surveillance video using a temporal convolutional 3D neural network (T-C3D). Then, a novel ranking loss function increases the distance between the classification scores of anomalous and normal videos, reducing the number of false negatives. The proposal has been evaluated and compared against state-of-art approaches, obtaining competitive performance without fine-tuning, which also validates its generalization capability. In this paper, the proposal design and reliability is presented and analyzed, as well as the aforementioned quantitative and qualitative evaluation in-the-wild scenarios, demonstrating its high sensitivity in anomaly detection in all of them.

Smart Video Surveillance System Based on Edge Computing.

New processing methods based on artificial intelligence (AI) and deep learning are replacing traditional computer vision algorithms. The more advanced systems can process huge amounts of data in large computing facilities. In contrast, this paper presents a smart video surveillance system executing AI algorithms in low power consumption embedded devices. The computer vision algorithm, typical for surveillance applications, aims to detect, count and track people's movements in the area. This application requires a distributed smart camera system. The proposed AI application allows detecting people in the surveillance area using a MobileNet-SSD architecture. In addition, using a robust Kalman filter bank, the algorithm can keep track of people in the video also providing people counting information. The detection results are excellent considering the constraints imposed on the process. The selected architecture for the edge node is based on a UpSquared2 device that includes a vision processor unit (VPU) capable of accelerating the AI CNN inference. The results section provides information about the image processing time when multiple video cameras are connected to the same edge node, people detection precision and recall curves, and the energy consumption of the system. The discussion of results shows the usefulness of deploying this smart camera node throughout a distributed surveillance system.

Comparing the Performance of Indoor Localization Systems through the EvAAL Framework.

In recent years, indoor localization systems have been the object of significant research activity and of growing interest for their great expected social impact and their impressive business potential. Application areas include tracking and navigation, activity monitoring, personalized advertising, Active and Assisted Living (AAL), traceability, Internet of Things (IoT) networks, and Home-land Security. In spite of the numerous research advances and the great industrial interest, no canned solutions have yet been defined. The diversity and heterogeneity of applications, scenarios, sensor and user requirements, make it difficult to create uniform solutions. From that diverse reality, a main problem is derived that consists in the lack of a consensus both in terms of the metrics and the procedures used to measure the performance of the different indoor localization and navigation proposals. This paper introduces the general lines of the EvAAL benchmarking framework, which is aimed at a fair comparison of indoor positioning systems through a challenging competition under complex, realistic conditions. To evaluate the framework capabilities, we show how it was used in the 2016 Indoor Positioning and Indoor Navigation (IPIN) Competition. The 2016 IPIN competition considered three different scenario dimensions, with a variety of use cases: (1) pedestrian versus robotic navigation, (2) smartphones versus custom hardware usage and (3) real-time positioning versus off-line post-processing. A total of four competition tracks were evaluated under the same EvAAL benchmark framework in order to validate its potential to become a standard for evaluating indoor localization solutions. The experience gained during the competition and feedback from track organizers and competitors showed that the EvAAL framework is flexible enough to successfully fit the very different tracks and appears adequate to compare indoor positioning systems.

Headgear Accessories Classification Using an Overhead Depth Sensor.

In this paper, we address the generation of semantic labels describing the headgear accessories carried out by people in a scene under surveillance, only using depth information obtained from a Time-of-Flight (ToF) camera placed in an overhead position. We propose a new method for headgear accessories classification based on the design of a robust processing strategy that includes the estimation of a meaningful feature vector that provides the relevant information about the people's head and shoulder areas. This paper includes a detailed description of the proposed algorithmic approach, and the results obtained in tests with persons with and without headgear accessories, and with different types of hats and caps. In order to evaluate the proposal, a wide experimental validation has been carried out on a fully labeled database (that has been made available to the scientific community), including a broad variety of people and headgear accessories. For the validation, three different levels of detail have been defined, considering a different number of classes: the first level only includes two classes (hat/cap, and no hat/cap), the second one considers three classes (hat, cap and no hat/cap), and the last one includes the full class set with the five classes (no hat/cap, cap, small size hat, medium size hat, and large size hat). The achieved performance is satisfactory in every case: the average classification rates for the first level reaches 95.25%, for the second one is 92.34%, and for the full class set equals 84.60%. In addition, the online stage processing time is 5.75 ms per frame in a standard PC, thus allowing for real-time operation.