The process of generating the cascade of classifiers is relatively time-consuming, but it produces quite promising results. The first three stages of a learned cascade are shown in Fig. . The time performance of the object detection crucially depends on the bootstrapping, i.e., on the generation of false positive examples during the stage learning. Nevertheless, learning has to be executed only once, the application of the cascade if very fast (300 ms). Thus the major time for the accurate object localization is spent during the model alignment and evaluation step (1.4 s).
The capabilities of the chosen approach have been evaluated in various experiments. Fig. shows four examples of successful detections and Table summarizes the object localization results.