To devise a typical application in this area involves three steps: detect moving objects, segment them and track them. In order to facilitate the development of the application, steps must be taken one at a time. This means that the easiest task is implemented first, in a simple context. Simple contexts include, among others, one static image produced to test a specific algorithm, static images taken from videos or pho- tographs with good conditions like color, contrast or scenario, a sequence of created images or video, and so forth. When one task is successfully performed under simple circumstances, it can be tested under real videos, starting by the videos with better quality or scenarios.In practice, this is not followed all the time for practical reasons. Usually it is required that a set of tasks be executed at a time, not only a single one. Thus, as soon as the first task produces acceptable results under acceptable conditions, the next step is started to be implemented. When all of the steps are implemented, another cycle takes place to further improve them, applying them in more realistic conditions.Additionally, it is important to use a framework, library or some toolkit to help the development of the application, avoiding to recode known algorithms and simplifying the design and implementation of the graphical user interface (GUI). Another important investment is to use an auxiliary tool to help the development and debugging of the application.B. Prototypical StudiesOpenCV has been chosen as the underlying application pro- gramming interface (API) for the development of our platform, basically due to the fact that it is a complete, widespread and popular library with a strong community support. One additional tool was also implemented to allow easy debugging and on-the-fly configuration of algorithms without requiring recompilation of the program. Fig. 4 shows the GUI of the testing framework developed.Despite being our original interest the characterization of traffic flow on aggregate basis, first steps taken toward the implementation of our application have shown some good potential for vehicle detection as well. Both optical flow, with or without pyramidal segmentation, and background sub- traction using the Gaussian background-foreground estimation model demonstrated promising results. However, these results still require improvements that might likely be accomplished by morphological operators or by improving the quality of the video. The use of contour and blob detection algorithms achieved good results that however declined with the increase of the density of vehicles, when occlusion begins to exist.