The principal motivation of this project is to minimize the number of deaths in car accidents. Passive safety systems (like airbags and belts) can decrease the number of fatal accidents and injures, while active safety systems can reduce the number of accidents. It is important not only to detect the vulnerable road users, but also to classify them. Distinction between various classification classes is done using theretical tools like monomial invariant features, support vector machine classifier and data fusion models.

Pedestrian recognition is a challenging task due to the variability of their appearance and poses. Moreover, the background of a traffic scene is incredibly unpredictable. Moreover, the background of a traffc scene is incredibly unpredictable, which makes the pedestrian segmentation especially diffcult. Various kinds of vehicle-based sensors are used to solve this task. Commonly used sensors are passive imaging sensors using visible light and infrared (IR) radiation, as well as active time-of-flight sensors, such as radar and lidar scanners. Imaging sensors are widely used because of their high lateral resolution and low cost, but extracting information from them involves substantial amount of processing. Furthermore, these sensors are very sensitive to the environment illumination and weather conditions. Time-of-flight sensors provide information about objects distances, but they do not deliver enough data to perform a complex classification. These two types of sensors complement each another, and their fusion is expected to present better results than single-sensor systems.

The goal of this project is to develop new methods for automatic detection and classification of road users on the basis of data fusion models and classification approaches. These methods should show reliable classification results and be computationaly efficient. They also should be appropriate for integration in the industry.