Scanning the Issue

Azim Eskandarian
2021 IEEE transactions on intelligent transportation systems (Print)  
Driver fatigue and stress significantly contribute to a higher number of car accidents worldwide. Although different detection approaches have already been commercialized and used by car producers (and third-party companies), research activities in this field are still needed in order to increase the reliability of these alert systems. This article presents a state-of-the-art review of different approaches for driver fatigue and stress detection and evaluation with the focus on various signals
more » ... biological, car, and video) and derived features used for these tasks. The description of different datasets, acquisition systems, and experiment scenarios are also included in this review. Vehicle and pedestrian detection are one of the critical tasks in autonomous driving. Since heterogeneous techniques have been proposed, the selection of a detection system with an appropriate balance among detection accuracy, speed and memory consumption for a specific task has become very challenging. To deal with this issue and to provide guidance for model selection, this article analyzes several mainstream object detection architectures, including faster R-CNN, R-FCN, and SSD, along with several typical feature extractors, such as ResNet50, ResNet101, MobileNet V1, MobileNet V2, inception V2, and inception ResNet V2. By conducting extensive experiments using the KITTI benchmark, which is a commonly used street dataset, the authors demonstrate that faster R-CNN ResNet50 obtains the best average precision (AP) (58%) for vehicle and pedestrian detection, with a speed of 8.6 f/s. Faster R-CNN inception V2 performs best for detecting cars and detecting pedestrians, respectively (74.5% and 47.3%). ResNet101 consumes the highest memory (9907 MB) and has the largest number of parameters (64.42 million), and inception ResNet V2 is the slowest model (3.05 f/s). SSD MobileNet V2 is the fastest model (70 f/s), and SSD MobileNet V1 is the lightest model in terms of memory usage (875 MB), both of which are suitable for applications on mobile and embedded devices. A volunteer assisted model of vehicular edge computing is proposed to utilize volunteer vehicles for handling the overloaded tasks in VEC servers. Volunteer vehicles are encouraged to assist the overloaded VEC servers via obtaining rewards from VEC servers. Based on a Stackelberg game, the interactions between requesting vehicles and VEC servers are analyzed, and the optimal strategies for them are found. Furthermore, the unique Stackelberg equilibrium in the game is proven theoretically, and a fast-searching algorithm based on a genetic algorithm is proposed to find the best pricing strategy for VEC servers. In addition, to maximize the reward of volunteer vehicles, the volunteer task assignment algorithm is proposed to get optimal mapping between the tasks and volunteer vehicles. Simulations are conducted to validate the effectivity of the proposed model and algorithms. A real-time high-performance deep convolutional neural network-based method is proposed for robust semantic segmentation of urban street scenes. Four key components, including a lightweight baseline network with atrous convolution and attention (LBN-AA), the distinctive atrous spatial pyramid pooling (DASPP), a spatial detail-preserving network (SPN), and a feature fusion network (FFN), are tightly combined and jointly optimized in an integrated network. The experimental results demonstrate that the proposed method offers excellent performance at the real-time speed for semantic segmentation of urban street scenes. Detection and tracking of moving objects (DATMO) in an urban environment using LiDAR is a major challenge for autonomous vehicles. The study presents a novel geometric model-free approach for DATMO using on-vehicle 2-D LiDAR. The approach depicts driving environment via static objects and moving objects using static obstacle map (SOM) and geometric model-free approach (GMFA), respectively. The interaction between GMFA and SOM estimates the correspondence between consecutive point clouds and utilize all point cloud without the region of interest in real-time. The proposed approach has been evaluated via RT range and labeled dataset. 1558-0016
doi:10.1109/tits.2021.3079675 fatcat:mtssvivuebcplifv6bw233jbkq