Método para mapeamento tridimensional térmico aplicado à sistemas de auxílio ao condutor

Abstract

Most traffic accidents are caused by human failure. Among these accidents, many involve pedestrians. One way to minimize this scenario is to use driver assistance systems. Such systems could act to prevent the accident if they receive reliable and timely information about the environment in which the vehicle is located. The use of information from the infrared spectrum provides a privileged view of the environment, as it is not subject to noise related to lighting, smoke and fog. In this sense, the present work intends to evaluate the estimation of pedestrian distances, based on a method capable of generating a three-dimensional thermal model, which can serve as input for a driver assistance system. This model consists of a series of steps, such as acquisition of thermal images, calibration, stereo matching, and matching optimization. The result of the method establishes a depth map capable of defining distances up to the observed objects. A genetic algorithm was used to optimize the parameters of the matching step of the method. The distances of pedestrians were evaluated in varied scenarios, and the root mean squared error metric was used to compare the values obtained with the real ones. The results obtained were satisfactory for distances up to 10m, where an error of 12.59% was obtained. Applying this method to driver assistance systems could prevent collisions with pedestrians in collector roads (40km/h) and minimize impact on local (30km/h) and arterial (60km/h) roads.