Forest fires are still a large concern in several countries due to the social, environmental and economic damages caused. This paper aims to show the design and validation of a proposed system for the classification of smoke columns with object detection and a deep learning-based approach. This approach is able to detect smoke columns visible below or above the horizon. During the dataset labelling, the smoke object was divided into three different classes, depending on its distance to the horizon, a cloud object was also added, along with images without annotations. A comparison between the use of RetinaNet and Faster R-CNN was also performed. Using an independent test set, an F1-score around 80%, a G-mean around 80% and a detection rate around 90% were achieved by the two best models: both were trained with the dataset labelled with three different smoke classes and with augmentation; Faster R-CNNN was the model architecture, re-trained during the same iterations but following different learning rate schedules. Finally, these models were tested in 24 smoke sequences of the public HPWREN dataset, with 6.3 min as the average time elapsed from the start of the fire compared to the first detection of a smoke column.

Drowsy driving is a prevalent and serious public health issue that deserves attention. Recent studies estimate that around 20% of car crashes have been caused by drowsy drivers. Nowadays, one of the main goals in the development of new advanced driver assistance systems is trustworthy drowsiness detection. In this paper, a drowsiness detection method based on changes in the respiratory signal is proposed. The respiratory signal, which has been obtained using an inductive plethysmography belt, has been processed in real time in order to classify the driver's state of alertness as drowsy or awake. The proposed algorithm is based on the analysis of the respiratory rate variability (RRV) in order to detect the fight against to fall asleep. Moreover, a method to provide a quality level of the respiratory signal is also proposed. Both methods have been combined to reduce false alarms due to the changes of measured RRV associated not with drowsiness but body movements. A driving simulator cabin has been used to perform the validation tests and external observers have rated the drivers' state of alertness in order to evaluate the algorithm performance. It has been achieved a specificity of 96.6%, a sensitivity of 90.3%, and Cohen's Kappa agreement score of 0.75 on average across all subjects through a leave-one-subject-out cross-validation. A novel algorithm for driver's state of alertness monitoring through the identification of the fight against to fall asleep has been validated. The proposed algorithm may be a valuable vehicle safety system to alert drowsiness while driving.

No abstract

This work proposes a systematic procedure to report the differences between heart rate variability time series obtained from alternative measurements reporting the spread and mean of the differences as well as the agreement between measuring procedures and quantifying how stationary, random and normal the differences between alternative measurements are. A description of the complete automatic procedure to obtain a differences time series (DTS) from two alternative methods, a proposal of a battery of statistical tests, and a set of statistical indicators to better describe the differences in RR interval estimation are also provided. Results show that the spread and agreement depend on the choice of alternative measurements and that the DTS cannot be considered generally as a white or as a normally distributed process. Nevertheless, in controlled measurements the DTS can be considered as a stationary process.

This work compares several fiducial points to detect the arrival of a new pulse in a photoplethysmographic signal using the built-in camera of smartphones or a photoplethysmograph. Also, an optimization process for the signal preprocessing stage has been done. Finally we characterize the error produced when we use the best cutoff frequencies and fiducial point for smartphones and photopletysmograph and compare if the error of smartphones can be reasonably be explained by variations in pulse transit time. The results have revealed that the peak of the first derivative and the minimum of the second derivative of the pulse wave have the lowest error. Moreover, for these points, high pass filtering the signal between 0.1 to 0.8 Hz and low pass around 2.7 Hz or 3.5 Hz are the best cutoff frequencies. Finally, the error in smartphones is slightly higher than in a photoplethysmograph.