Drowsiness has become one of the major causes of road accidents now-a-days. In order to alleviate this issue, a system has been developed, which uses electroencephalogram (EEG) signals to detect drowsiness with sufficient reliability. This experiment was conducted on a small population and the EEG signals were acquired using a 14-channel wireless headset, while they were in a virtual driving environment. To extract the eye closures, the EEG signal was segmented, and pre-processed. Further the scalograms which describes the time-frequency characteristics of these segments were taken. Pretrained Convolutional Neural Network based architectures viz. ResNet-152, ResNet101, VGG16, VGG19, AlexNet were used to distinguish three states of the driver namely “Sleepy or Drowsy”, “Asleep” and “Awake”.

A Comparative Study of Drowsiness Detection From Eeg Signals Using Pretrained CNN Models

Drowsiness poses a serious challenge to road safety and various in-cabin sensing technologies have been experimented with to monitor driver alertness. Cameras offer a convenient means for contactless sensing, but they may violate user privacy and require complex algorithms to accommodate user (e.g., sunglasses) and environmental (e.g., lighting conditions) constraints. This paper presents a lightweight convolution neural network that measures eye closure based on eye images captured by a wearable glass prototype, which features a hot mirror-based design that allows the camera to be installed on the glass temples. The experimental results showed that the wearable glass prototype, with the neural network in its core, was highly effective in detecting eye blinks. The blink rate derived from the glass output was highly consistent with an industrial gold standard EyeLink eye-tracker. As eye blink characteristics are sensitive measures of driver drowsiness, the glass prototype and the lightweight neural network presented in this paper would provide a computationally efficient yet viable solution for real-world applications.

/pdf/a-cnn-based-wearable-system-for-driver-drowsiness-detection-1zcudwv7.pdf

A CNN-Based Wearable System for Driver Drowsiness Detection

5G or ZTE the latest improvement to the existing 4G communication standard. These technologies could be evaluated by various metrics called performance indicators among which throughput plays a major role. Throughput is the measure of the rate of data transferred to the device. Higher the throughput, better is the performance of the network. This work models and analyses the throughput obtained with the variations observed on the identified parameters on which it depends on. Here the problem is analysed as a regression problem and hence regressor models are applied. Multiple models ranging from statistical to probabilistic and machine learning to deep recurrent networks are analysed with a 10 fold cross validation. Also, the effect of dimensionality reduction is applied to the dataset and the performance is observed. It is noticed from the work that the top performing models are consistent in performance measured using the regression metrics. 

Throughput Analysis with Effect of Dimensionality Reduction on 5G Dataset using Machine Learning and Deep Learning Models

People must incorporate a “work from home” strategy because of the COVID-19 outbreak. In today's pandemic situation, due to working from home, employees are working for long hours, and spending long hours is a pretty challenging task. Nowadays, irrespective of age concerns, sciatica, Kyphosis, and lordosis are becoming a significant problem even for youngsters. The longest nerve in our body is sciatica, which causes severe pain due to stress applied while sitting in the wrong posture. It gets compressed with our lower back discs, which may lead to severe radiating pain from our lower back disc to the entire right leg, and a person can't even perform his daily activities comfortably. To prevent these problems, sitting posture while working should be maintained correctly. This work mainly focuses on preventing employers, and students from sciatica, Kyphosis, and Lordosis health issues. We used all kinds of sitting postures that interact while working with a laptop, classified which posture was good, and predicted which stance led to health issues. We used Convolutional Neural Network and K-Nearest Neighbor machine learning algorithms to predict the correct sitting postures. In KNN, we followed two techniques to improve the performance: using Edge detection. The other method we used was detecting facial landmark detection and plotting their respective rotational angles. So by using this technique, we improved the accuracy and precision rate compared to the classical Edge detection. We also trained the model with CNN, which gives good results. We performed a comparative analysis to pick the best model to integrate with OpenCV to make it real-time. 

Comparative Analysis of Machine Learning and Deep Learning Algorithms for Real-Time Posture Detection to Prevent Sciatica, Kyphosis, Lordosis

5G or fifth generation is the latest in the communication technology which is being researched worldwide as a successor to the current 4G technology. 5G operates on higher bandwidth with higher data rates of the order of Gbit/s. 5G is estimated to play a major role in the development of smart cities and IoT use cases. Lumos 5G is one of the groups researching on the topic. In this paper, the throughput obtained under various conditions is analysed as a regression model in machine learning with the features as continuous variables. It is observed that the newer tree machine learning models are performing better on the dataset than the traditional tree models. This is verified by performing a tenfold cross-validation check on the best performing models. 

Modelling 5G Data Using Tree-Based Machine Learning Models

Forewarning drowsy drivers can reduce the number of road accidents. A non-intrusive drowsiness detection system is implemented, which alerts the driver on the onset of drowsiness. A Pi camera module attached to Raspberry Pi is used to acquire and process the live video of the driver. Haar face detector in OpenCV is used for face detection followed by 68 points of facial landmark identification. Eye and Mouth Aspect Ratios, blink rate and yawning rate are the features extracted. Drowsiness detection is done using two methodologies viz. a threshold-based one and the other, employing artificial intelligence. The machine learning techniques used are LDA and SVM. Feedback is provided as an alarm if a driver is found to be drowsy. The analysis shows that machine learning-based techniques viz. LDA and SVM outperform threshold technique for the dataset considered.

Radheswarreddy Madireddy

Papers

Driver Drowsiness Detection System Using Conventional Machine Learning