Over the last years deep learning methods have been shown to outperform previous state-of-the-art machine learning techniques in several fields, with computer vision being one of the most prominent cases. This review paper provides a brief overview of some of the most significant deep learning schemes used in computer vision problems, that is, Convolutional Neural Networks, Deep Boltzmann Machines and Deep Belief Networks, and Stacked Denoising Autoencoders. A brief account of their history, structure, advantages, and limitations is given, followed by a description of their applications in various computer vision tasks, such as object detection, face recognition, action and activity recognition, and human pose estimation. Finally, a brief overview is given of future directions in designing deep learning schemes for computer vision problems and the challenges involved therein.

/pdf/deep-learning-for-computer-vision-a-brief-review-31uavl0qvq.pdf

Deep Learning for Computer Vision: A Brief Review.

Driver decisions and behaviors are essential factors that can affect the driving safety. To understand the driver behaviors, a driver activities recognition system is designed based on the deep convolutional neural networks (CNN) in this paper. Specifically, seven common driving activities are identified, which are the normal driving, right mirror checking, rear mirror checking, left mirror checking, using in-vehicle radio device, texting, and answering the mobile phone, respectively. Among these activities, the first four are regarded as normal driving tasks, while the rest three are classified into the distraction group. The experimental images are collected using a low-cost camera, and ten drivers are involved in the naturalistic data collection. The raw images are segmented using the Gaussian mixture model to extract the driver body from the background before training the behavior recognition CNN model. To reduce the training cost, transfer learning method is applied to fine tune the pre-trained CNN models. Three different pre-trained CNN models, namely, AlexNet, GoogLeNet, and ResNet50 are adopted and evaluated. The detection results for the seven tasks achieved an average of 81.6% accuracy using the AlexNet, 78.6% and 74.9% accuracy using the GoogLeNet and ResNet50, respectively. Then, the CNN models are trained for the binary classification task and identify whether the driver is being distracted or not. The binary detection rate achieved 91.4% accuracy, which shows the advantages of using the proposed deep learning approach. Finally, the real-world application are analyzed and discussed.

/pdf/driver-activity-recognition-for-intelligent-vehicles-a-deep-3b5cabqg07.pdf

Driver Activity Recognition for Intelligent Vehicles: A Deep Learning Approach

A review of Convolutional-Neural-Network-based action recognition

Multi-sensor information fusion based on machine learning for real applications in human activity recognition: State-of-the-art and research challenges

The feasibility of real-time drowsiness detection using commercially available, off-the-shelf, lightweight, wearable electroencephalogram (EEG) sensors is explored. While EEG signals are known to be reliable indicators of fatigue and drowsiness, they have not been used widely due to their size and form factor. However, the use of lightweight wearable EEGs alleviates this concern. Spectral analysis of EEG signals from these sensors using support vector machines (SVMs) is shown to classify drowsy states with high accuracy. The system is validated using data collected on 23 subjects in fresh and drowsy states. An accuracy of 81% is obtained at a per-subject level and 74% in cross-subject validation using SVM with radial basis kernel. Using a temporal aggregation strategy, the cross-subject validation accuracy is shown to improve to 87%. The EEG signals are also used to characterise the blink duration and frequency of subjects. However, classification of drowsy states using blink analysis is shown to have lower accuracy than that using spectral analysis.

Real-time drowsiness detection using wearable, lightweight brain sensing headbands

Convolutional neural networks (ConvNets) coupled with long short term memory (LSTM) networks have been recently shown to be effective for video classification as they combine the automatic feature extraction capabilities of a neural network with additional memory in the temporal domain. This paper shows how multiview fusion can be applied to such a ConvNet LSTM architecture. Two different fusion techniques are presented. The system is first evaluated in the context of a driver activity recognition system using data collected in a multicamera driving simulator. These results show significant improvement in accuracy with multiview fusion and also show that deep learning performs better than a traditional approach using spatiotemporal features even without requiring any background subtraction. The system is also validated on another publicly available multiview action recognition dataset that has 12 action classes and 8 camera views.

Vlad Kecojevic

Papers

Real-time drowsiness detection using wearable, lightweight brain sensing headbands

Multiview fusion for activity recognition using deep neural networks