Pedestrian detection is used in many vision based applications ranging from video surveillance to autonomous driving. Despite achieving high performance, it is still largely unknown how well existing detectors generalize to unseen data. This is important because a practical detector should be ready to use in various scenarios in applications. To this end, we conduct a comprehensive study in this paper, using a general principle of direct cross-dataset evaluation. Through this study, we find that existing state-of-the-art pedestrian detectors, though perform quite well when trained and tested on the same dataset, generalize poorly in cross dataset evaluation. We demonstrate that there are two reasons for this trend. Firstly, their designs (e.g. anchor settings) may be biased towards popular benchmarks in the traditional single-dataset training and test pipeline, but as a result largely limit their generalization capability. Secondly, the training source is generally not dense in pedestrians and diverse in scenarios. Under direct cross-dataset evaluation, surprisingly, we find that a general purpose object detector, without pedestrian-tailored adaptation in design, generalizes much better compared to existing state-of-the-art pedestrian detectors. Furthermore, we illustrate that diverse and dense datasets, collected by crawling the web, serve to be an efficient source of pre-training for pedestrian detection. Accordingly, we propose a progressive training pipeline and find that it works well for autonomous-driving oriented pedestrian detection. Consequently, the study conducted in this paper suggests that more emphasis should be put on cross-dataset evaluation for the future design of generalizable pedestrian detectors. Code and models can be accessed at this https URL.

Generalizable Pedestrian Detection: The Elephant In The Room

Pedestrian detection is used in many vision based applications ranging from video surveillance to autonomous driving. Despite achieving high performance, it is still largely unknown how well existing detectors generalize to unseen data. This is important because a practical detector should be ready to use in various scenarios in applications. To this end, we conduct a comprehensive study in this paper, using a general principle of direct cross-dataset evaluation. Through this study, we find that existing state-of-the-art pedestrian detectors, though perform quite well when trained and tested on the same dataset, generalize poorly in cross dataset evaluation. We demonstrate that there are two reasons for this trend. Firstly, their designs (e.g. anchor settings) may be biased towards popular benchmarks in the traditional single-dataset training and test pipeline, but as a result largely limit their generalization capability. Secondly, the training source is generally not dense in pedestrians and diverse in scenarios. Under direct cross-dataset evaluation, surprisingly, we find that a general purpose object detector, without pedestrian-tailored adaptation in design, generalizes much better compared to existing state-of-the-art pedestrian detectors. Furthermore, we illustrate that diverse and dense datasets, collected by crawling the web, serve to be an efficient source of pre-training for pedestrian detection. Accordingly, we propose a progressive training pipeline and find that it works well for autonomous-driving oriented pedestrian detection. Consequently, the study conducted in this paper suggests that more emphasis should be put on cross-dataset evaluation for the future design of generalizable pedestrian detectors. Code and models can be accessed at https://github.com/hasanirtiza/Pedestron.

/pdf/generalizable-pedestrian-detection-the-elephant-in-the-room-48xy8n39up.pdf

Over the past decades Machine Learning and Deep Learning algorithm played a vital part in the development of Autonomous Vehicle. It is indeed for the perception system to examine the environment around the vehicle and identify the objects such as pedestrian, vehicle and traffic signals, etc. Using this information, control system module can take necessary action to control the vehicle in terms of braking, speed, lane change or steering, etc. This paper focuses on the survey of machine learning algorithms and techniques applied in the design of autonomous driving system over a decade. Performance of each algorithm was analyzed in terms of prediction time and accuracy have been documented and compared.

A Comprehensive Survey on Autonomous Driving Cars: A Perspective View

The ability to perceive and understand surrounding road-users behaviors is crucial for self-driving vehicles to correctly plan reliable reactions. Computer vision that relies mostly on machine learning techniques enables autonomous vehicles to perform several required tasks such as pedestrian detection. Furthermore, within a fully autonomous driving environment, driverless vehicle has to communicate and share perceived data with its neighboring vehicles for more safe navigation. In this context, our paper proposes a warning notification diffusion solution related to real-time pedestrian presence detection, through an inter-vehicle communication system. To achieve this purpose, pedestrian and vehicle recognition is required. Thus, we implemented intended detectors. We used Histogram of Oriented Gradients (HOG) descriptor with the linear Support Vector Machine (SVM) classifier for the pedestrian detector, and Haar feature-based cascade classifier to reach vehicle detection. The performance evaluation of our solution leads to fairly good detection accuracy around 90% for pedestrian and 88% for vehicle.

Pedestrian Detection for Autonomous Driving within Cooperative Communication System

In recent years, the network technology known as Internet of Vehicles (IoV) has been developed to improve road safety and vehicle security, with the goal of servicing the digital demands of car drivers and passengers. However, the highly dynamical network topology that characterizes these networks, and which often leads to discontinuous transmissions, is one of the most significant challenges of IoV. To address this issue, IoV infrastructure-based components known as roadside units (RSU) are designed to play a critical role by providing continuous transmission coverage and permanent connectivity. However, the main challenges that arise when deploying RSUs are balancing IoVs’ performances and total cost so that optimal vehicle service coverage is provided with respect to some target Quality of Service (QoS) such as: service coverage, throughput, low latency, or energy consumption. This paper provides an in-depth survey of RSU deployment in IoV networks, discussing recent research trends in this field, and summarizing of a number of previous papers on the subject. Furthermore, we highlight that two classes of RSU deployment can be found in the literature—static and dynamic—the latter being based on vehicle mobility. A comparison between the existing RSU deployment schemes proposed in existing literature, as well as the various networking metrics, are presented and discussed. Our comparative study confirms that the performance of the different RSU placement solutions heavily depends on several factors such as road shape, particularity of road segments (like accident-prone ones), wireless access methods, mobility model, and vehicles’ distribution over time and space. Besides that, we review the most important RSU placement approaches, highlighting their strengths and limitations. Finally, this survey concludes by presenting some future research directions in this domain.

/pdf/roadside-unit-deployment-in-internet-of-vehicles-systems-a-3gcbkyim.pdf

Roadside Unit Deployment in Internet of Vehicles Systems: A Survey

In recent years, the emergence of the Internet of Vehicles (IoV) aims to enhance the users’ quality of experience through proposing more sophisticated services ranging from guaranteeing the user safety to improving his comfort. The IoV ecosystem is complex, heterogeneous, and evolving. Many entities participate to compose its architecture (such as vehicles, humans, roadside units, ITS). Moreover, different communication types co-exist to ensure the IoV connectivity and continuity. This diversity leads to new security requirements that seem more complex to take into account and enlarge the attack surface of such ecosystem. Many security mechanisms should be considered to enforce the security of IoV environment at many levels: data, entities, communications, storage, etc.. Trust management is one of the potential security mechanisms that aims to increase the reliability within the IoV environment. The topic has been widely explored within the vehicular ad hoc networks (VANETs). However, the VANET represents only one component of the IoV ecosystem. Thus, the approaches proposed for the VANET should be adapted to be applied for the IoV. Moreover, the advent of the emerging technologies like blockchain, cloud, SDN as well as artificial intelligence bring new opportunities to propose more relevant approaches within the trust management mechanisms within the IoV context. Accordingly, this survey deals with the literature about the trust management topic in vehicular environments. The scope considers the IoV environment as well as the relevant approaches proposed for the VANET context since it is one of the important component of the IoV ecosystem. We start by quickly reviewing the existing surveys about security of the vehicular environments. Then, we give a general overview about trust concepts. Afterwards, we present the security and trust challenges and attacks in the vehicular context. Later, we classify and compare the most relevant approaches related to the trust management for the IoV proposing a new taxonomy. We complete this survey by highlighting the open future directions and perspectives for research. 

A survey of trust management in the Internet of Vehicles

Vehicular networking is an enabling technology that promises a huge advancement in the field of Intelligent Transport System (ITS)and mobile networks. Vehicular Ad hoc NetworK (VANET) is emerging quickly as a new architecture that is based on V2X communications. Vehicles are equipped with Electronic Control Units (ECUs) which are interconnected via various intra-vehicle networks (e.g., LIN, CAN, Ethernet, MOST, and FlexRay). Accordingly, a communication gateway is required to facilitate interfacing between the different vehicular environment components. To the best of our knowledge, contributions in this area are very limited. In this context, this paper proposes a new In-car gateway architecture for the Intra and Inter vehicular networks in order to perform the required specifications for V2X communications.

In-car Gateway Architecture for Intra and Inter-vehicular Networks

The Internet of Vehicles (IoV) is an Internet of Things (IoT) application that offers several utilities such as traffic analysis, safe driving, road optimization, and travel comfort. Software-Defined Networking (SDN) technology has been shown to provide various benefits to support the IoV. However, the construction of IoV makes it a complex system posing several challenges among which the important ones are security and privacy of data. Intrusion Detection Systems (IDSs) have been proposed in the IoV to identify cyber attacks and protect private data. Recently work has started to implement IDSs based on Federated learning as collaborative IDSs have proved effective security of IoV. In another hand, trust management has revolutionized the IoV filed, providing decision-making support to secure the network. Stating that an SDN-driven IoV architecture in which nodes trustworthiness gets assessed can provide a promising framework for IDS, we propose in this paper a Federated learning-based IDS for the IoV under the SDN structure. We integrate trust metrics to assist in securing the IoV network. Simulation experiments are conducted to validate the proposal.

Federated learning based IDS approach for the IoV

Transportation is a vastly developed research field that tries to optimize humans' safety in everyday life. Recently, autonomous vehicles are being thoroughly developed by a lot of companies such as Uber[1] and Tesla[2], which will improve safety on the roads. These vehicles are using artificial intelligence to correctly interact with their environment. First, we developed a Java program called “ARIBAN” which provides algorithms for recognition and classification of traffic signs with Multi-Layer Perceptron Neural Networks (MLPNN). Second, we use back propagation algorithm in a supervised manner to establish the network. Third, the neural networks are trained and used with a large amount of traffic signs. Finally, the post-processing combines the results to make a recognition decision. Eventually, we have tested our trained network with more than 62 types of traffic signs. Experimental results have demonstrated the effectiveness of the proposed system. Furthermore, the proposed system was deployed within an architecture for autonomous driving.

Amal Hbaieb

Papers

A survey of trust management in the Internet of Vehicles

Pedestrian Detection for Autonomous Driving within Cooperative Communication System

In-car Gateway Architecture for Intra and Inter-vehicular Networks

Federated learning based IDS approach for the IoV

Traffic Sign Recognition Using Neural Networks Useful for Autonomous Vehicles