Object detection and recognition are an important issue in the field of Computer Vision, where its accomplishment in both real time and low false positives rates has became the main goal of various research works, including the ones related to new interactivity forms in Digital TV. This dissertation proposes a software system based on machine learning that allows an efficient training for new objects and performs their subsequent recognition in real time, for both static images and digital videos. The proposed system is based on the use of Haar features of the object, which require a low computation time for their calculation, and on the usage of a cascade of classifiers, which allows a quick discard of image areas that does not contain the desired object while having a low occurrence of false positives. Through the use of image segmentation techniques, the system turns the search for objects into an extremely fast operation in high-resolution videos. Furthermore, through the use of parallelism techniques, one can simultaneously detect various objects without losing performance.