Vision-based gait impairment analysis for aided diagnosis

Gait is a firsthand reflection of health condition. This belief has inspired recent research efforts to automate the analysis of pathological gait, in order to assist physicians in decision-making. However, most of these efforts rely on gait descriptions which are difficult to understand by humans, or on sensing technologies hardly available in ambulatory services. This paper proposes a number of semantic and normalized gait features computed from a single video acquired by a low-cost sensor. Far from being conventional spatio-temporal descriptors, features are aimed at quantifying gait impairment, such as gait asymmetry from several perspectives or falling risk. They were designed to be invariant to frame rate and image size, allowing cross-platform comparisons. Experiments were formulated in terms of two databases. A well-known general-purpose gait dataset is used to establish normal references for features, while a new database, introduced in this work, provides samples under eight different walking styles: one normal and seven impaired patterns. A number of statistical studies were carried out to prove the sensitivity of features at measuring the expected pathologies, providing enough evidence about their accuracy. Graphical Abstract Graphical abstract reflecting main contributions of the manuscript: at the top, a robust, semantic and easy-to-interpret feature set to describe impaired gait patterns; at the bottom, a new dataset consisting of video-recordings of a number of volunteers simulating different patterns of pathological gait, where features were statistically assessed. Graphical abstract reflecting main contributions of the manuscript: at the top, a robust, semantic and easy-to-interpret feature set to describe impaired gait patterns; at the bottom, a new dataset consisting of video-recordings of a number of volunteers simulating different patterns of pathological gait, where features were statistically assessed.


Introduction
ical emergencies in hospital environments [22]. These 21 results are supported by different sensors for extract- 22 ing gait data, being wearable gadgets and vision-based 23 devices those most popular. Sensors in the first group 24 (e.g., gyroscopes, accelerometers, markers) [11,13] ac-25 quire precise information, although they can be deemed ing highly accurate motion data without requiring any 32 contact with a sensor [1]. However, they are generally 33 costly and demand certain setting and calibration pro-34 cesses, hence their use tends to be restricted to more 35 specialized environments. On the contrary, less sophisgait from healthy people have been recently published. 83 In [10], a wearable 2D system based on an smartphone 84 fixed in a belt is proposed. The phone includes a cam-85 era which tracks two markers placed on feet to com-86 pute step lenght, width and time, gait speed and double 87 support time. In another work [24], a simple RGB we-88 bcam is used together with markers to get kinematic 89 gait parameters from people walking in a treadmill. 90 Concurrently, 3D low-cost approaches have gained in 91 popularity since Microsoft Kinect was released. For in-92 stance, in [3] and [4] a Kinect-based marker-less solution 93 was validated against a more sophisticated system con-       Fig. 1 Gait cycle from the right limb perspective through its phases stance and swing. Events heel strike (HS) and toe off (TO) determine the start and end of these phases. The complementary stance/swing distribution for the opposite limb is also included in the lower part. This image is inspired in one from [29].
These theoretical assumptions are considered nec-237 essary conditions for normal gait, but not sufficient.

238
That is, a pathological gait can potentially yield identi-    asymmetry measure A f can be defined as follows:   Note that both measures do not depend on frame rate.

348
Conventionally, detection of start and end of these 349 phases is carried out by identifying the HS and TO 350 events within gait cycles [7,19]. Nevertheless, patholog-             Table 1 One-sample t-tests given a known population mean for stance phase (StP ) and swing phase (SwP ) features over the nm sequences from INIT Gait Database. Symbols "•" highlight p-values above the significance level α = 0.05, indicating irrelevant differences between the sample and the population theoretical mean.    Table 2 Unpaired two-sample t-tests assuming equal variances between neutral sequences from INIT Gait Database and OU-ISIR Database. Features involved are the asymmetries in step length (A Sl ), intensity (A I ) and amplitude (A Am ), and the fall risk factor (F r). Symbols "•" highlight p-values above the significance level α = 0.05, indicating irrelevant differences between both samples.   Table 3 Paired two-sample t-tests performed on the INIT Gait Database between neutral (nm) sequences and full body affected (fb) sequences. Symbols "•" ("•") highlight p-values above (below) the significance level α = 0.05, indicating irrelevant (substantial) differences between samples.

597
The upper half of Table 3 shows the results of paired  The t-test results corresponding to the six involved 620 features are shown in Table 4. By way of summary, in    As regards the second study (Section 3.2), Table 3 661 shows consistent behaviors of the primary features when  Table 4 Paired two-sample t-tests performed on the INIT Gait Database between neutral (nm) sequences and right leg half motion (l-r0.5 ) or left leg half motion (l-l0.5 ) sequences. Symbols "•" ("•") highlight p-values above (below) the significance level α = 0.05, indicating irrelevant (substantial) differences between samples.   915 Javier Ortells has recently received a PhD in Computer Science from the University Jaume I, Spain. His thesis addressed gait analysis for both biometric recognition and medical diagnosis purposes.
M. Trinidad Herrero-Ezquerro has a PhD in Medicine and Surgery from the University of Navarra, Spain. She works as a full professor at the University of Murcia, Spain, and she is currently the Director of the Institute for Research on Aging, at the University of Murcia. Her main research interests are clinical and experimental neuroscience, movement disorders, and neurodegenerative diseases.
Ramón A. Mollineda has a PhD in Computer Science from the Polytechnic University of Valencia, Spain. He is currently an Associate Professor in the Department of Computer Languages and Systems at the University Jaume I, Spain. His research interests are machine learning, computer vision, biometrics, and gait analysis