Search

Visual sensing of the environment is crucial for flying an unmanned aerial vehicle (UAV) and is a centerpiece of many related applications. The ability to run computer vision and machine learning algorithms onboard an unmanned aerial system (UAS) is becoming more of a necessity in an effort to alleviate the communication burden of high-resolution video streaming, to provide flying aids, such as obstacle avoidance and automated landing, and to create autonomous machines. Thus, there is a growing interest on the part of many researchers in developing and validating solutions that are suitable for deployment on a UAV system by following the general trend of edge processing and airborne computing, which transforms UAVs from moving sensors into intelligent nodes that are capable of local processing. In this paper, we present, in a rigorous way, the design and implementation of a 12.85 kg UAV system equipped with the necessary computational power and sensors to serve as a testbed for image processing and machine learning applications, explain the rationale behind our decisions, highlight selected implementation details, and showcase the usefulness of our system by providing an example of how a sample computer vision application can be deployed on our platform. [ABSTRACT FROM AUTHOR]

This work introduces the design, architecture, implementation, and testing of a low-cost and machine-learning-enabled device to be worn on the wrist. The suggested wearable device has been developed for use during emergency incidents of large passenger ship evacuations, and enables the real-time monitoring of the passengers' physiological state, and stress detection. Based on a properly preprocessed PPG signal, the device provides essential biometric data (pulse rate and oxygen saturation level) and an efficient unimodal machine learning pipeline. The stress detecting machine learning pipeline is based on ultra-short-term pulse rate variability, and has been successfully integrated into the microcontroller of the developed embedded device. As a result, the presented smart wristband is able to provide real-time stress detection. The stress detection system has been trained with the use of the publicly available WESAD dataset, and its performance has been tested through a two-stage process. Initially, evaluation of the lightweight machine learning pipeline on a previously unseen subset of the WESAD dataset was performed, reaching an accuracy score equal to 91%. Subsequently, external validation was conducted, through a dedicated laboratory study of 15 volunteers subjected to well-acknowledged cognitive stressors while wearing the smart wristband, which yielded an accuracy score equal to 76%. [ABSTRACT FROM AUTHOR]

Nowadays, the transformations of cities into smart cities is a crucial factor in improving the living conditions of the inhabitants as well as addressing emergency situations under the concept of public safety and property loss. In this context, many sensing systems have been designed and developed that provide fire detection and gas leakage alerts. On the other hand, new technologies such edge computing have gained significant attention in recent years. Moreover, the development of recent intelligent applications in IoT aims to integrate several types of systems with automated next-generation emergency calls in case of a serious accident. Currently, there is a lack of studies that combine all the aforementioned technologies. The proposed smart building sensor system, SB112, combines a small-size multisensor-based (temperature, humidity, smoke, flame, CO, LPG, and CNG) scheme with an open-source edge computing framework and automated Next Generation (NG) 112 emergency call functionality. It involves crucial actors such as IoT devices, a Public Safety Answering Point (PSAP), the middleware of a smart city platform, and relevant operators in an end-to-end manner for real-world scenarios. To verify the utility and functionality of the proposed system, a representative end-to-end experiment was performed, publishing raw measurements from sensors as well as a fire alert in real time and with low latency (average latency of 32 ms) to the middleware of a smart city platform. Once the fire was detected, a fully automatic NG112 emergency call to a PSAP was performed. The proposed methodology highlights the potential of the SΒ112 system for exploitation by decision-makers or city authorities. [ABSTRACT FROM AUTHOR]