This paper investigates the classification of driving styles using unsupervised learning techniques applied to recorded driving data. The study focuses on identifying two primary driving styles: calm and aggressive. The importance of lane change scenarios in discriminating between these styles is highlighted, using features such as lateral speeds and yaw angles. Using spectral clustering and K-means algorithms, a driving style detection method is proposed. The obtained results indicate that K-means outperforms spectral clustering in effectively classifying drivers based on their behaviour, particularly in lane change situations. This research contributes to a deeper understanding of driver behaviour on the road and provides insights into the potential applications of unsupervised learning in driving style recognition.

High blood pressure (HBP) is at the root of many cardiovascular diseases, causing millions of deaths every year worldwide. Continuous monitoring of blood pressure (BP) is necessary, not only to prevent cardiovascular disease, but also to help those already affected. While many methods of blood pressure measurement exist at present, they present two major inconve-niences, these methods do not allow continuous measurement and they are frequently very invasive. In this paper, we present an alternative method to measure systolic and diastolic blood (SBP and DBP) pressure values with photoplethysmographic (PPG) signals using a deep learning architecture. PPG signals come from MIMIC II, a noisy and unprocessed database. Using a ResNet neural network coupled with LSTM layers, we succeeded in predicting SBP and DBP values, with an RMSE of 8.96 mmHg.

This review discusses the importance of detecting driver emotions to improve driving safety and user experience. The article presents recent literature on emotion recognition in the context of driving, reviewing different models of emotion representation, recent public databases for driver emotion recognition, and various uni-modal and multi-modal methods to detect driver emotions. The study shows that detecting the driver's emotional state and its intensity is vital to improving driving safety and the user experience, particularly if the emotion is disconnected from or not related to the driving task.

The expectations regarding autonomous vehicles are very high to transform the future mobility and ensure more road safety. Autonomous driving system should be able in the short term to detect dangerous situations and respond appropriately and thus increase driving safety. Understanding the intentions of drivers has recently received growing interest. A long-term prediction method based on gated unit-recurrent neural network model is proposed for the problem of trajectory prediction of surrounding vehicles. A deep neural network with Long-short term memory (LSTM) and Gated Recurrent Units (GRU) structure is used to analyze the spatial-temporal features of the past trajectory. Through sequences learning, the system generates the future trajectory of other traffic participants for different horizons of prediction. We evaluate all models with standard metric (Root mean square error RMSE), loss function convergence and processing time. After comparing the different models, our experiments revealed that the proposed GRU based models is indeed better than LSTM based models in term of accuracy and processing speed.

We present and investigate, in the first part of this paper, a set of published works that may be employed to remotely quantify mental stress based on physiological signals by non-contact means. Several techniques can be used to this purpose, thermal imaging being currently the most advanced in our knowledge. Webcams correspond to a ubiquitous and to the most accessible techniques in the particular purpose of mental stress detection. After all these theoretical reminders, we present a pilot study based on a new framework that was developed to detect mental workload changes using video frames obtained from a low-cost webcam. To induce stress, we have employed a computerized Stroop color word test on twelve subjects. The results offer further support for the applicability of mental workload detection by remote and low-cost means, providing an alternative to conventional contact techniques.

Generally, Negative emotions can lead to health problems. In order to detect negative emotions, an advanced method of the EMG signal analysis is presented. Negative emotions of interest in this work are: fear, disgust and sadness. These emotions are induced with presentation of IAPS (International Affective Picture System) images. The EMG signal is chosen to extract a set of characteristic parameters to be used for classification of emotions. The analysis of EMG signal is performed using the wavelet transform technique to extract characteristic parameters while the classification is performed using the SYM (Separator Vector Machine) technique. The results show a good recognition rates using these characteristic parameters.

In this paper, we focus our attention on the development of a virtual reality exposure system (VRE) to induce anxiety to phobic people and on a strategy for recognizing it. We describe a short term anxiety detection from blood volume pulse (BVP) measurement, we detail data collection, feature extraction and classification. We propose a model of anxiety detection using support vector machines (SVM) and evaluate it on the collected data. Results show that our model detects anxiety with a good accuracy. This study aims to support the psychologist in social anxiety diagnosis.

The ability to recognize emotion is one of the hallmarks of emotion intelligence. This paper proposed to recognize emotion using physiological signals obtained from multiple subjects. IAPS images were used to elicit target emotions. Five physiological signals: Blood volume pulse (BVP), Electromyography (EMG), Skin Conductance (SC), Skin Temperature (SKT) and Respiration (RESP) were selected to extract 30 features for recognition. Two pattern classification methods, Fisher discriminant and SVM method are used and compared for emotional state classification. The experimental results indicate that the proposed method provides very stable and successful emotional classification performance as 92% over six emotional states.

In this paper, we present a detection and tracking feature points algorithm in real time camera input environment. To trace and extract a face image, we use a modified face detector based on the Haar-like features. For feature points detection, we use good features to track of Shi and Thomasi. In order to track the facial feature points, pyramidal Lucas-Kanade feature tracker algorithm is used. Results on the real time indicate that the proposed algorithm can accurately extract facial features points.

2D Zernike moments belong to the useful object invariant descriptors which have been successfully applied in pattern recognition tasks. The main problem of using Zernike moments invariants is that they are not able to discriminate two objects having the same shape. In this paper, an approach based on Zernike moments applied on color images is proposed. A support vector machine is used for object classification. For object segmentation, the connected component labeling algorithm is used. Compared with the classical method, this approach shows higher accuracy in object recognition. Some experimental results on the COIL-100 database are presented.

In this paper, we propose a new approach for some augmented reality applications by exploiting minimal geometric knowledge and using Desargues theorem. The idea underlying our approach is to use a generalization of Desargues theorem in uncalibrated images context. This approach allows the realization of three applications that includes: points matching, novel view synthesis and adding a virtual object to real scene. Examples on real and synthetic images are presented.