Abstract:
People Re-Identification has become a topic of in terest due to the increasing use of intelligent video surveillance systems in the security industry. In this paper, we implement a people Re-Id system comprising three important modules: a) a person detection, responsible for detecting people in the image, b) preprocessing module, responsible of extracting the soft biometric features of the detected persons, and c) an identification module, capable of identifying the detected person. For this purpose, a two branches multi-input and one output network model is built. The first one receives the body silhouette descriptor and the other the clothing texture descriptor. To train this model a dataset of 7 identities was built, with 1862 and 481 images for training and validation respectively, facing problems such as the existing bias in the public datasets. In addition, two videos and one validation image set were used to evaluate the system performance. The results of our proposal are positive, demonstrating that the combination of soft-biometrics features, body silhouette and clothing textures of the person increases the system ability to Re-Identify a person in images an qd videos
