UNIVERSIDAD AUTÓNOMA DE MADRID ESCUELA POLITÉCNICA SUPERIOR people detection algorithms based on appearance and motion information

.The work presented in this document is focused on the improvement of the people detection task in visual analysis in video surveillance sequences. We have designed and implemented a people detection algorithm based on appearance and motion as discriminative information. Firstly, we have undertaken a study on the state of the art in people detection algorithms targeted to video surveillance scenarios and we have made a classi cation of the di erent algorithms studied. Within this classi cation, algorithms based on appearance use to obtain better results than motion based algorithms, and therefore we have designed and implement an algorithm based on appearance but adding robustness to the detection with motion information. As main result a new people detector have been implemented and integrated into the VPULab system: this new algorithm obtains more reliable results and less computational cost than the previous algorithm implemented by VPULab. Additionally, in order to provide a good performance evaluation of the proposed framework and a comparison with previous work: the design and annotation of a people video/image dataset have been done.