An approach for gait anonymization using deep learning

Abstract

The human gait has become another biometric trait used in security systems because it is unique to each person and can be recognized at a distance. However, a bad actor could use a gait recognition system to identify a person on the basis of his or her gait. We have developed a gait anonymization method that prevents unauthorized gait recognition. It modifies the gait so that the person cannot be identified while maintaining the naturalness of the gait. The modification is done by adding another gait, called "noise gait". A convolutional neural network makes this modification by taking two gaits as input, the original gait and the noise gait, and outputting an anonymized gait. The proposed method was evaluated using the success rate and mean opinion score (MOS). The success rate is the rate of failed gait recognition, and the MOS is a measure of the naturalness of the anonymized gait. In our experiments, the success rate achieved 98.86% at most while the highest naturalness score is 3.73 in the MOS scale. These findings should open new research directions regarding privacy protection related to gait recognition.

Spatio-temporal generative adversarial network for gait anonymization

“…Step 1 corresponds to Step 1 in the method of Tieu et al [1], but the input here is a color video instead of a binary video. Likewise, Step 3 corresponds to Step 3 in the method of Tieu et al; however, the binary anonymized gaits are colorized to obtain color anonymized gaits, as explained in Subsection E. Our two main contributions are found…”

“…in Step 2. The success of anonymization is improved by using random noise instead of a noise gait as previously used [1]. The naturalness of the anonymized gait is improved by using an ST-GAN containing a generator and two discriminators (a spatial discriminator and a temporal discriminator).…”

“…We evaluated the performance of our method by using the CASIA-B dataset [13] and two metrics suggested by Tieu et al [1], the success rate and naturalness. We used two gait recognition systems as black-box systems for measuring the success rate, one by Zheng et al [14] and one by Wu et al [15].…”

“…The human gait, i.e., the manner and pattern of walking, has become an important biometric trait because it is unique to each person and can be recognized at a distance without physical contact or the person's cooperation [1]. However, a serious privacy problem may arise if the person in a video is identified unintentionally by a gait recognition system because his/her personal information (e.g., name, address, occupation) may eventually be revealed.…”

“…These methods generally focus on privacy protection and do not address the problem of retaining gait naturalness. As far as we know, only one reported method [1] is aimed at anonymizing gaits so that they cannot be recognized while retaining their naturalness. The model used by this method anonymizes a gait by using a deep neural network to add a noise gait to the original one.…”