Resumen de Advanced driver assistance system based on computer vision using detection, recognition and tracking of road signs
Los accidentes de tráfico son un grave problema socioeconómico. Obviamente el coste humano es imposible de evaluar y el económico supone un continuo e ingente gasto de dinero por parte de los gobiernos. Se han propuesto diferentes soluciones para paliar los efectos de los accidentes, una de las cuales, los Sistemas Avanzados de Ayuda a la Conducción, son el marco en el que se encuadra el presente trabajo. Estos sistemas, como su nombre indica, asisten al conductor ofreciéndole información del entorno o actuando en determinadas circunstancias para la salvaguarda de los ocupantes del vehículo, o para facilitar la conducción. El sistema que se propone en esta tesis es una plataforma multipropósito original en su concepción, cuyo fin más inmediato es reconocer las señales de tráfico de prohibición, peligro, ceda el paso, obligación e indicación. La información obtenida de ese reconocimiento se integra dentro de un módulo de aviso al conductor. Lo que se pretende es que el conductor conozca en todo momento si está contraviniendo alguna norma de circulación derivada de una velocidad o maniobra inadecuada para el tipo de señal que se ha reconocido. Dado que este sistema está embarcado en un vehículo, deberá cumplir dos requisitos de especial importancia: funcionar en tiempo real y tanto en entorno urbano como en autopista. El primero cobra especial relevancia si se piensa en que la seguridad de los ocupantes del vehículo y de los peatones puede depender de los avisos que permitan al conductor anticiparse a un peligro. La segunda, que es una aportación original, garantiza que el sistema funcionará en vías donde la velocidad es mayor y por tanto también la probabilidad y gravedad de un posible accidente. El sistema, como se decía anteriormente, sirve al desarrollo de otras aplicaciones, como es el caso del inventariado automático de señales de tráfico, tan en auge actualmente. ______________________________________________ Road traffic accidents are a serious socio-economic problem, where the cost of human life is impossible to evaluate, and cause massive and continuous government spending. Different solutions have been proposed to reduce the effects of accidents, one of which, Advanced Driver Assistance Systems, forms part of the framework which encompasses the current investigation work presented in this thesis. These systems, as their name suggest, assist the driver by providing vital information on the traffic environment or by acting under speciffic circumstances to safeguard the occupants of the vehicle, or to facilitate driving. A multitask driver assistance platform is originally presented as part of the research work of this thesis, among other tasks, it has been designed to recognize road signs in both urban and non-urban environments. The road signs that have been considered are: prohibition, danger, yield, obligation and indication. The information obtained from the road sign recognition process forms part of a complete module that advises drivers on traffic circulation requirements. The primary objective of this work has been to increase driver awareness, at all times, on the legal limitations which have established for road safety. The two areas which have been considered in this thesis are the velocity of the vehicle and the velocity of the vehicle corresponding to incorrect driving manoeuvres both of which are controlled using the information contained within road signs. The assistance platform which has been designed forms an integral part of the vehicle, thus it must satisfy two important requirements: first, provide the driver with real time information from road signs and secondly, to operate in both urban and non-urban environments. Real time information is important for the safety of drivers, passengers, and pedestrians where the information provided warns the driver well in advance of any danger so that the appropriate manoeuvres can be made to correct the speed of the vehicle. One important aspect of the work presented here is that the system has also been designed for non-urban environments, such as: national roads, toll roads and motorways where there is a higher probability of more serious and fatal accidents occurring due to the increased speed. There is a wide range of possible applications for road sign recognition systems, another area of interest which has motivated the work carried out for this thesis has been an automatic road sign inventory system. From the beginning of research in automatic road sign detection applications, many di®erent stages have been proposed, such as: signal detection, road sign recognition, and road sign tracking. The work presented in this thesis provides an in depth analysis of each of these three stages and this has allowed a more robust and complete system to be designed. In this thesis an exhaustive review is presented on color spaces and their characteristic color components which are best suited to the task of searching for road signs within an image. The technique of template matching using patterns with road signs has been optimized in this research, also included is an analysis of the most adequate models required to effciently detect each type of road sign. As part of the development of the recognition stage of the system, the two currently most important tools used in object recognition have been studied, these are: template matching and neural networks (NN). A comparative analysis of both of these techniques has been performed, where emphasis has been placed on image preprocessing to optimize the results. The final stage of this system addresses the problem of road sign tracking. The method proposed is a model based on the movement of the camera within the vehicle with respect to the road sign which is taken as the reference point. All the different stages of the system, which have been developed, form part of an experimental platform used on-board a test vehicle. To investigate the viability of this system the experimental trials have been carried out under real conditions. This methodology has been used for each stage, and the results presented corroborate the advantages and effectiveness of a multitask driver assistance platform.
