fabriciotorquato/person-re-identification

When they have an environment monitored by security cameras where they need to determine if a person who is appearing in the image was already present in the video before or if it is their first appearance, it is called within the field of artificial intelligence research as re-identifying people. This problem is presented as a challenge among researchers since such images collected by security cameras can present various visual noises in addition to partial obstructions by people. Thus, inspired by the teaching of the Greek historian Herodotus, thinking about the past to understand the present and idealizing the future, this work is presented through metrics between FaceNet and MobileNetV2 when exposed to the WiseNet dataset facing the problem of re-identification, being that both networks had very close F-Measure performances and accuracy, however when compared to the inference time of the experiment and the training time, FaceNet shows its advantages, in which with these inferred data it hoped to have helped to contribute to future evolutions and improvements in the algorithms of the area.

Python