Gait Recognition: Databases, Representations, and Applications

Abstract

There has been considerable progress in automatic recognition of people by the way they walk since its inception almost 20 years ago: there is now a plethora of technique and data which continue to show that a person's walking is indeed unique. Gait recognition is a behavioural biometric which is available even at a distance from a camera when other biometrics may be occluded, obscured or suffering from insufficient image resolution (e.g. a blurred face image or a face image occluded by mask). Since gait recognition does not require subject cooperation due to its non-invasive capturing process, it is expected to be applied for criminal investigation from CCTV footages in public and private spaces. This article introduces current progress, a research background, and basic approaches for gait recognition in the first three sections, and two important aspects of gait recognition, the gait databases and gait feature representations are described in the following sections.Publicly available gait databases are essential for benchmarking individual approaches, and such databases should contain a sufficient number of subjects as well as covariate factors to realize statistically reliable performance evaluation and also robust gait recognition. Gait recognition researchers have therefore built such useful gait databases which incorporate subject diversities and/or rich covariate factors.Gait feature representation is also an important aspect for effective and efficient gait recognition. We describe the two main approaches to representation: model-free (appearance-based) approaches and model-based approaches. In particular, silhouette-based model-free approaches predominate in recent studies and many have been proposed and are described in detail.Performance evaluation results of such recent gait feature representations on two of the publicly available gait databases are reported: USF Human ID with rich covariate factors such as views, surface, bag, shoes, time elapse; and OU-ISIR LP with more than 4,000 subjects. Since gait recognition is suitable for criminal investigation applications of the gait recognition to forensics are addressed with real criminal cases in the application section. Finally, several open problems of the gait recognition are discussed to show future research avenues of the gait recognition.

The OU-ISIR Gait Database comprising the Large Population Dataset with Age and performance evaluation of age estimation

“…This database is suitable for the evaluation of which covariate significantly impacts gait recognition performance [47] and can be used for the analysis of gait-based age estimation. However, the limited number of subjects, and significant biases in terms of gender and age, may make the evaluation of age estimation inadequate.…”

Gait recognition using normalized shadows

“…Therefore, a GTI possesses the symmetry property expected from TILT inputs. When dealing with shadow silhouettes, the GTI can be obtained by vertically flipping the binary images containing the shadow silhouette and averaging all the resulting binary images ( , , ) belonging to the entire gait sequence, according to equation (1).…”

“…A more limited set of biometric traits can be used to perform recognition over a distance and without user cooperation. A popular example among them is gait [1]. In the literature, gait recognition is performed following either a model-based, or an appearance-based approach.…”

Gait Recognition

“…According to the overview [37] about the gait dataset, most of datasets are based on 2D videos or based on 3D motion data captured by professional camera, such as VICON. To our best knowledge, there are a few gait datasets containing both 2D silhouette images and 3D joints position information.…”

“…Here, the Euclidean distance is chosen to measure the space distance referring to the research experiences in Refs. [37,38].…”