Anonymized person re-identification in surveillance cameras

Person re-identification (Re-ID) is a valuable technique because it can assist in finding suspects after a terrorist attack. However, the machine learning algorithms for person Re-ID are usually trained on large datasets with images of many different people in a public space. This could pose privacy concerns for the people involved. One way to alleviate this concern is to anonymize the people in the dataset. Anonymization is important to minimize the storage and processing of personal information, such as facial information in a surveillance video. However, anonymization typically leads to loss of information and could lead to severe deterioration of the Re-ID quality. In this paper, we show that it is possible to store only anonymized person detections while still achieving a high quality person Re-ID. This leads to the conclusion that for the development of re-identification algorithms in situations where privacy is of great importance it is not necessary to store facial information in person re-identification datasets.