Abstract

Traffic sign recognition is important for the development of driver assistance systems and fully autonomous vehicles. Even though GPS navigator systems works well for most of the time, there will always be situations when they fail. In these cases, robust vision based systems are required. Traffic signs are designed to have distinct colored fields separated by sharp boundaries. We propose to use locally segmented contours combined with an implicit star-shaped object model as prototypes for the different sign classes. The contours are described by Fourier descriptors. Matching of a query image to the sign prototype database is done by exhaustive search. This is done efficiently by using the correlation based matching scheme for Fourier descriptors and a fast cascaded matching scheme for enforcing the spatial requirements. We demonstrated on a publicly available database state of the art performance. "p"Original Publication: Fredrik Larsson and Michael Felsberg, Using Fourier Descriptors and Spatial Models for Traffic Sign Recognition, SCIA konferens, 23-27 May 2011, Ystad Sweden, 2011, Lecture Notes in Computer Science, Image Analysis, 238-249. http://dx.doi.org/10.1007/978-3-642-21227-7_23 Copyright: Springer Document type: Part of book or chapter of book

Abstract

The use of Computer Vision techniques for the automatic recognition of road signs is fundamental for the development of intelligent vehicles and advanced driver assistance systems. In this paper, we describe a procedure based on color segmentation, Histogram of Oriented Gradients (HOG), and Convolutional Neural Networks (CNN) for detecting and classifying road signs. Detection is speeded up by a preprocessing step to reduce the search space, while classification is carried out by using a Deep Learning technique. A quantitative evaluation of the proposed approach has been conducted on the well-known German Traffic Sign data set and on the novel Data set of Italian Traffic Signs (DITS), which is publicly available and contains challenging sequences captured in adverse weather conditions and in an urban scenario at night-time. Experimental results demonstrate the effectiveness of the proposed approach in terms of both classification accuracy and computational speed. Document type: Part of book or chapter of book

Abstract

The automatic generation of street networks is attracting the attention of research and industry communities in areas such as routable map generation. This paper presents a novel mechanism that focuses on the automatic detection of street elements such as traffic lights, street crossings and roundabouts which could be used to generate street maps and populate them with traffic influencing infrastructural elements such as traffic lights. In order to minimize the system requirements and simplify the data collection from many users with minimal impact for them, only traces of GPS data from a mobile device while driving are used. Speed and acceleration time series are derived from the GPS data. An outlier detection algorithm is used first in order to detect abnormal driving locations (which can be due to infrastructural elements or particular traffic conditions). Using deep learning, speed and acceleration patterns are automatically analyzed at each outlier in order to extract relevant features which are then classified into a traffic light, street crossing, urban roundabout or other element. The classification results are enhanced by adding the degree of atypicity for each point calculated as the percentage of times that a particular location is detected as an outlier in several drives. The proposed algorithm achieves a combined recall of 0.89 and a combined precision of 0.88 for classification. The research leading to these results has received funding from the “HERMES-SMART DRIVER” project TIN2013-46801-C4-2-R (MINECO), funded by the Spanish Agencia Estatal de Investigación (AEI), and the “ANALYTICS USING SENSOR DATA FOR FLATCITY” project TIN2016-77158-C4-1-R (MINECO/ ERDF, EU) funded by the Spanish Agencia Estatal de Investigación (AEI) and the European Regional Development Fund (ERDF). Document type: Article

Abstract

Co-encadrement de la thèse : Bogdan Stanciulescu; Pedestrian Detection and Traffic Sign Recognition (TSR) are important components of an Advanced Driver Assistance System (ADAS). This thesis presents two methods for eliminating false alarms in pedestrian detection applications and a novel three stage approach for TSR. Our TSR approch consists of a color segmentation, a shape detection and a content classification phase. The red color enhancement is improved by using an adaptive threshold. The performance of the K-d tree is augmented by introducing a spatial weighting. The Random Forests yield a classification accuracy of 97% on the German Traffic Sign Recognition Benchmark. Moreover, the processing and memory requirements are reduced by employing a feature space reduction. The classifiers attain an equally high classification rate using only a fraction of the feature dimension, selected using the Random Forest or Fisher's Criterion. This technique is also validated on two different multiclass benchmarks: ETH80 and Caltech 101. Further, in a static camera video surveillance application, the immobile false positives, such as trees and poles, are eliminated using the correlation measure over several frames. The recurring false alarms in the pedestrian detection in the scope of an embedded ADAS application are removed using a complementary tree filter.; La détection de piétons et la reconnaissance des panneaux routiers sont des fonctions importantes des systèmes d'aide à la conduite (anglais : Advanced Driver Assistance System - ADAS). Une nouvelle approche pour la reconnaissance des panneaux et deux méthodes d'élimination de fausses alarmes dans des applications de détection de piétons sont présentées dans cette thèse. Notre approche de reconnaissance de panneaux consiste en trois phases: une segmentation de couleurs, une détection de formes et une classification du contenu. Le color enhancement des régions rouges est amélioré en introduisant un seuil adaptatif. Dans la phase de classification, la performance du K-d tree est augmentée en utilisant un poids spatial. Les Random Forests obtiennent un taux de classification de 97% sur le benchmark allemand de la reconnaissance des panneaux routiers (German Traffic Sign Recognition Benchmark). Les besoins en mémoire et calcul sont réduits en employant une réduction de la dimension des caractéristiques. Les classifieurs atteignent un taux de classification aussi haut qu'avec une fraction de la dimension des caractéristiques, selectionée en utilisant des Random Forests ou Fisher's Crtierion. Cette technique est validée sur deux benchmarks d'images multiclasses : ETH80 et Caltech 101. Dans une application de vidéo surveillance avec des caméras statiques, les fausses alarmes des objets fixes, comme les arbres et les lampadaires, sont éliminées avec la corrélation sur plusieurs trames. Les fausses alarmes récurrentes sont supprimées par un filtre complémentaire en forme d'arbre.