A major concern in large cities is to minimize the effects of the increasing quantity of vehicles in circulation and, consequently, of the accidents that tend to occur more frequently. Due to the popularization and miniaturization of GPS receivers, the availability of large volumes of data about vehicle speed in urban roads and the large number of traffic-related messages published in social networks, it is now possible to collect enough input data to model traffic conditions based on the observed reduction in speed values. However, it is necessary to filter the data to minimize thematic, spatial and temporal uncertainties. This thesis proposes a methodology to assess the impact of traffic accidents by analyzing speed values. To achieve this goal, it also proposes auxiliary methodologies, aiming at: (1) processing GPS-tracked routes to compute speed statistics and estimate traffic in two-way streets, by performing direction analysis; (2) representing traffic behavior based on the observed speed values; (3) extracting and selecting accident-related data by mining Twitter posts for later identification of the likely effects on speed values. The main contributions of this thesis are: (1) the assessment of traffic conditions based on speed values, which are easier to acquire than data about traffic volume and concentration; (2) the use of posts from social networks, which provide timely access to traffic events; (3) the assessment of urban roads instead of freeways or roads, which require modeling intersections, traffic lights and pedestrian flow; and (4) a methodology designed to extract speed statistics from raw GPS data, which handles likely error sources related to both map matching process and temporal classification.