Object detection exists in many countries around the world after a recent growing interest for autonomous vehicles in the last decade. This paper focuses on a vision-based approach focusing on vehicles and pedestrians detection in real-time as a perception for autonomous vehicles, using a convolutional neural network for object detection. A developed YOLOv3-tiny model is trained with the INRIA dataset to detect vehicles and pedestrians, and the model also measures the distance to the detected objects. The machine learning process is leveraged to describe each step of the training process, it also combats overfitting and increases the speed and accuracy. The authors were able to increase the mean average precision; a way to measure accuracy for object detectors; 31.3\% to 62.14\% based on the result of the training that was done. Whilst maintaining a speed of 18 frames per second.

Cette thèse s'inscrit dans le cadre de la conception de véhicules autonomes, et plus spécifiquement de la vérification de contrôleurs de tels véhicules. Nos contributions à la résolution de ce problème sont les suivantes : (1) fournir une syntaxe et une sémantique pour un modèle de systèmes hybrides, (2) étendre les fonctionnalités du model checker statistique Cosmos à ce modèle et (3) valider empiriquement la pertinence de notre approche sur des cas d'étude typiques du véhicule autonome.Nous avons choisi de combiner le modèle des réseaux de Petri stochastiques de haut niveau (qui était le formalisme d'entrée de Cosmos) avec le formalisme d'entrée de Simulink afin d'atteindre un pouvoir d'expression suffisant. En effet Simulink est très largement utilisé dans le domaine automobile et de nombreux contrôleurs sont spécifiés avec cet outil. Or Simulink n'a pas de sémantique formellement définie. Ceci nous a conduit à concevoir une telle sémantique en deux temps : tout d'abord en introduisant une sémantique dite exacte mais qui n'est pas opérationnelle puis en la complétant par une sémantique approchée intégrant le facteur d'approximation recherché.Afin de combiner le modèle à événements discrets des réseaux de Petri et le modèle continu spécifié en Simulink, nous avons proposé au niveau syntaxique une interfacereposant sur de nouveaux types de transitions et au niveau sémantique une extension de la boucle de simulation. L'évaluation de ce nouveau formalisme a été entièrement implémentée dans Cosmos.Grace à ce nouveau formalisme, nous avons développé et étudié les deux cas d'étude suivants : d'une part une circulation dense sur une section d'autoroute et d'autre part l'insertion du véhicule dans une voie rapide. L'analyse des modélisations correspondantes a démontré la pertinence de notre approche. / This thesis takes place in the context of autonomous vehicle design, and concerns more specifically the verification of controllers of such vehicles. Our contributions are the following: (1) give a syntax and a semantics for a hybrid system model, (2) extend the capacities of the model-checker Cosmos to that kind of models, and (3) empirically confirm the relevance of our approach on typical case studies handling autonomous vehicles.We chose to combine high-level stochastic Petri nets (which is the input formalism of Cosmos) with the input formalism of Simulink, to obtain an adequate expressive power. Indeed, Simulink is largely used in the automotive industry and numerous controllers have been specified using this tool. However, there is no formal semantics for Simulink, which lead us to define such a semantics in two steps:first, we propose an exact (but not operational) semantics, then we complete it by an approximate semantics that includes the targeted approximation level.In order to combine the discrete event model of Petri nets and the continous model specified in Simulink, we define a syntactic interface that relies on new transition types; its semantics consists of an extension of the simulation loop. The evaluation of this new formalism has been entirely implemented into Cosmos.Using this new formalism, we have designed and studied the two following case studies: on one hand, a heavy traffic on a motorway segment, and on the other hand the insertion of a vehicle into a motorway. Our approach has been validated by the analysis of the corresponding models.

Indiana University-Purdue University Indianapolis (IUPUI) / This thesis addresses a critical issue of automotive safety. As traffic is increasing on the roads day by day, road safety is also a very important concern. Driving simulators can play an extensive role in the development and testing of advanced safety systems in peculiar traffic environments, respectively. Advanced Driver Assist Systems (ADAS) are getting enormous reputation but there is still need for more improvements. This thesis presents a design of an Automatic Emergency Pullover (AEP) strategy using active safety systems for a semi-autonomous vehicle. The idea for this system is that a moving vehicle equipped with an AEP system can automatically pull over on the roadside safely when the driver is considered incapable of driving. Furthermore, AEP supporting features such as; Lane Keeping Assist, Blind Spot Monitoring, Vehicle and Pedestrian Automatic Emergency Braking, Adaptive Cruise Control are also included in this work. The designs for application of each system have been explained along with its algorithms, model development, component architecture, simulation results, vehicular/pedestrian behavior and trajectory precision on software tools provided by Realtime Technologies, Inc. All major variables which influence the performance of vehicle after AEP activation, have been observed and remodeled according to control algorithms. The implementation of AEP system which can control vehicle dynamics has been verified with the help of simulation results.

Indiana University-Purdue University Indianapolis (IUPUI) / For pedestrians and autonomous vehicles (AVs) to co-exist harmoniously and safely in the real-world, AVs will need to not only react to pedestrian actions, but also anticipate their intentions. In this thesis, we propose to use rich visual and pedestrian-environment interaction features to improve pedestrian crossing intention prediction from the ego-view.We do so by combining visual feature extraction, graph modeling of scene objects and their relationships, and feature encoding as comprehensive inputs for an LSTM encoder-decoder network. Pedestrians react and make decisions based on their surrounding environment, and the behaviors of other road users around them. The human-human social relationship has al-ready been explored for pedestrian trajectory prediction from the bird’s eye view in stationary cameras. However, context and pedestrian-environment relationships are often missing incurrent research into pedestrian trajectory, and intention prediction from the ego-view. To map the pedestrian’s relationship to its surrounding objects we use a star graph with the pedestrian in the center connected to all other road objects/agents in the scene. The pedestrian and road objects/agents are represented in the graph through visual features extracted using state of the art deep learning algorithms. We use graph convolutional networks, and graph autoencoders to encode the star graphs in a lower dimension. Using the graph en-codings, pedestrian bounding boxes, and human pose estimation, we propose a novel model that predicts pedestrian crossing intention using not only the pedestrian’s action behaviors(bounding box and pose estimation), but also their relationship to their environment. Through tuning hyperparameters, and experimenting with different graph convolutions for our graph autoencoder, we are able to improve on the state of the art results. Our context-driven method is able to outperform current state of the art results on benchmark datasetPedestrian Intention Estimation (PIE). The state of the art is able to predict pedestrian crossing intention with a balanced accuracy (to account for dataset imbalance) score of 0.61, while our best performing model has a balanced accuracy score of 0.79. Our model especially outperforms in no crossing intention scenarios with an F1 score of 0.56 compared to the state of the art’s score of 0.36. Additionally, we also experiment with training the state of the art model and our model to predict pedestrian crossing action, and intention jointly. While jointly predicting crossing action does not help improve crossing intention prediction, it is an important distinction to make between predicting crossing action versus intention.

No description available.

No description available.

The perception system is the key for autonomous vehicles to sense and understand the surrounding environment. As the cheapest and most mature sensor, monocular cameras create a rich and accurate visual representation of the world. The objective of this thesis is to investigate if camera-based deep learning models with transfer learning technique can achieve 2D object detection, License Plate Detection and Recognition (LPDR), and highway lane detection in real time. The You Only Look Once version 3 (YOLOv3) algorithm with and without transfer learning is applied on the Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) dataset for cars, cyclists, and pedestrians detection. This application shows that objects could be detected in real time and the transfer learning boosts the detection performance. The Convolutional Recurrent Neural Network (CRNN) algorithm with a pre-trained model is applied on multiple License Plate (LP) datasets for real-time LP recognition. The optimized model is then used to recognize Ontario LPs and achieves high accuracy. The Efficient Residual Factorized ConvNet (ERFNet) algorithm with transfer learning and a cubic spline model are modified and implemented on the TuSimple dataset for lane segmentation and interpolation. The detection performance and speed are comparable with other state-of-the-art algorithms. / Thesis / Master of Applied Science (MASc)

Liability for damages caused by an autonomous vehicle Abstract For several years now, autonomous vehicles have been one of the most interesting topics associated with the upcoming Industry 4.0 and the spread of artificial intelligence in society. This thesis therefore deals with the topic of autonomous vehicles as a subset of autonomous systems, specifically in connection with civil-law liability. It is probable that in the future there will be a massive expansion of autonomous means of transport in society, and it cannot be ruled out that a situation will arise where their setting will result in damage. The aim of this thesis is to present the models of liability, the use of which is discussed by jurisprudence in connection with autonomous vehicles, and to assess the possibility of using the current institutes of Civil Code, to find a suitable future solution. For this purpose, the thesis first deals with the concept of autonomous vehicles according to the Civil Code and other legislation and by using existing definitions of artificial intelligence deals with specific features of autonomous systems, as well as the status of specific persons associated with the operation of autonomous vehicles in light of current legislation, their typology and conditions of their operation itself. The second part of this...

Recently, various self-driving and driving assistance systems such as Advanced Driver Assistance System (ADAS) have been developed with the intent to reduce the number of motor vehicle accidents. While self-driving systems have been proven to reduce traffic accidents, the systems sometimes make other drivers confused because of their mechanical behavior. To avoid confusion and possible error, it is necessary to construct self-driving systems that exhibit human-like behaviors. Risk Potential theory has been used to construct models that successfully represent driver behavior, especially expert behavior. This project uses Risk Potential theory to construct and evaluate a collision avoidance driver model which uses braking to avoid potential collisions with pedestrians. As a first step, a basic driver model which uses Risk Potential theory is constructed and evaluated using metrics such as collision avoidance, comfortability, and false alarm avoidance. Second, human driving data is collected to observe driver’s risk perception during interactions with a pedestrian. Finally, our proposed driver models improve on standard RP model’s performance but comparisons of the models with observed human performance reveal opportunities for further improvement.

Autonomous vehicles are developing at a rapid pace while pedestrians' experience with autonomous vehicles is less researched. This paper reported an exploratory study where 40 participants encountered a conditionally autonomous vehicle with unusual driver behaviors at crossing by watching videos and photos. Questionnaires and semi-structured interviews were used to investigate pedestrians' experience. The results showed distracted driver behaviors in the conditionally autonomous driving context had negative impact on pedestrians' crossing experience. Black window on conditionally autonomous vehicles made pedestrians feel uncomfortable and worried. / Autonoma fordon utvecklas i snabb takt medan fotgängares erfarenhet av autonoma fordon är mindre undersökt. I denna uppsats redovisades en undersökande studie där 40 deltagare observerat ett villkorligt autonomt fordon med ovanliga förarbeteenden vid en korsning, genom att titta på videor och foton. Frågeformulär och semi-strukturerade intervjuer användes för att undersöka fotgängares erfarenhet. Resultaten visade att distraherade förarbeteenden i det villkorliga autonoma förhållandet hade negativ inverkan på fotgängares upplevelse vid övergångsstället. Svarta vindrutor på villkorligt autonoma fordon gör att fotgängare känner sig obekväma och oroliga.