THEORETICAL ADVANCES An aperiodic feature representation for gait recognition in cross-view scenarios for unconstrained biometrics Chandrashekhar Padole 1 • Hugo Proenc ¸a 1 Received: 20 June 2014 / Accepted: 15 March 2015 Ó Springer-Verlag London 2015 Abstract The state-of-the-art gait recognition algorithms require a gait cycle estimation before the feature extraction and are classified as periodic algorithms. Their effective- ness substantially decreases due to errors in detecting gait cycles, which are likely to occur in data acquired in non- controlled conditions. Hence, the main contributions of this paper are: (1) propose an aperiodic gait recognition strat- egy, where features are extracted without the concept of gait cycle, in case of multi-view scenario; (2) propose the fusion of the different feature subspaces of aperiodic fea- ture representations at score level in cross-view scenarios. The experiments were performed with widely known CASIA Gait database B, which enabled us to draw the following major conclusions, (1) for multi-view scenarios, features extracted from gait sequences of varying length have as much discriminating power as traditional periodic features; (2) for cross-view scenarios, we observed an av- erage improvement of 22 % over the error rates of state-of- the-art algorithms, due to the proposed fusion scheme. Keywords Gait representation  Multi-view gait  Cross- view gait  Aperiodic gait recognition  Gait cycle estimation  Unconstrained biometrics 1 Introduction There is an increasing interest in the human gait to be employed in biometrics applications. It is more suitable to be used in less controlled scenarios [19], character- ized by the reduced quality of data. Gait recognition is an activity-based biometric trait [10], and represents the subjects in the way they walk. Even if the discriminating capability across humans gaits is smaller than that of classical biometric traits (e.g., the iris, or the face), there are several reasons for using the gait as a biometric trait that can be pointed out here: (1) no minutia information is used in the recognition process, leading easier to ac- quire the data from long distances; (2) walking is an instinctive activity of humans, reducing the possibility to imitations or deliberate changes over a large period; and (3) unlike the face and iris, a gait information is easily captured from multiple view angles, hence, reduces the possibility of having significant occlusions in a gait sequence. A view-dependent gait recognition, where gallery and probe samples are of same view angle, makes subject registration process complex and infeasible. On one side, it restricts the probe subject to walk in a particular direction, in which gallery data were acquired and other side, it re- quires to have gallery data acquired in all possible view angles. It also suffers from being sensitive to the difference between angles of probe and gallery data. To overcome these problems, the gait recognition research has ap- proached towards the direction of view-independent recognition methods, which are more suitable for uncon- strained biometrics. Further, view-independent methods can be applied in two different scenarios, namely, multi- view and cross-view gait recognitions [15], which are de- tailed in next Sect. 2. & Chandrashekhar Padole chandupadole@yahoo.com Hugo Proenc ¸a hugomcp@di.ubi.pt 1 Department of Computer Science, IT, Instituto de Telecomunicac ¸o ˜es, University of Beira Interior, 6200 Covilha ˜, Portugal 123 Pattern Anal Applic DOI 10.1007/s10044-015-0468-0