Gait Pattern Recognition And Detection Computer Science Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Gait refers as the way and style human walking. Scientists' researches in field of ethology and psychology show that human gait could be unique. The interests of analyses gait as new emerging biometrics also get stronger now.


In general, most of us have the experience that we can recognize a familiar friend by their walking. Although the factors which help to recognize them are not only by the way of their walking but also include other reasons. For example, their clothing, hair style etc. Cutting and Kozlowski's research [3] provided that such human gait recognition can identify of friends. After Cutting and Kozlowski's work, there all sorts of experiment show that people can recognize gender due to the cue of gait. Psychological researches [1, 2] also showed human gait derived from video can be trusted evidence to identify individual. These findings inspired experts of computer vision to derived human gait pattern from images and video. Study of human gait, as well as develop gait as a new emerging biometric method for people identification, is gaining increasing attractions now. Human gait pattern as a newly biometric also has its own natural advantages compared with other various biometrics, such as iris, fingerprint, hand geometry, face and voice etc. The most impressive advantage when using gait as biometric identification is that it can recognize people at a certain distance with a low-resolution video, where one object only represents on small number of pixels.

Currently there is much research in analyzing human gait via computer vision techniques. From spatiotemporal patterns method by Niyogi [4] in 1994 which generates a characteristic pattern in a spatiotemporal (XYT) volume. The various analyses of gait techniques evolve continually. Mainly there are two branches of those analyses, one is shape-based analysis and the other is model-based analysis. Shape-based analyses focus on the silhouette of the typical human object. These kind of analyses obtain the human silhouette by derive the moving object from the image and video background [7]. After that, the individual human object can be recognized by measurements that the person's movement or shape. There is also model-based analysis aim to derive the torso and the legs movement of human body. Unlike the shape-based analysis, this method mainly focuses on body dynamics or body in motion. For example the relative motion of the angles between the limbs or the dynamic change value on the hip and leg rotation.

In the process of human gait recognition, we derive numerical information from video images to reflect the identity of moving object. This paper researches the possibility of identify people by their gait pattern and the way to derive their gait signature by a novel model-based approach - a three points model. The three points of this triangle model each represent the human body's head and two feet. This paper looks for an analysis to represent the walking path drawn out by this triangle human model, model it to polygon and then identify human object in video.

The remainder of this paper is organized as follows. First it reviews various current biometrics and discuss the advantage and possibility of using human gait pattern as a biometrics. Then in Section 3 describes the current approaches to gait recognition. Here represent nearly all the significant human gait recognition approaches from 1994 to 2010 and classify those approaches by their features. At the end of this paper concluded the appendix.


Gait is the pattern of moving of the limbs of human (or animals), include walking and running. Walking is the most effective and convenient method for people travel in a short distance, since walking is the body's most natural means of the movement. Walking simply is a repeatable cycle of alternative strike of the right and left heel, as the Fig.3 show a typical single man walking cycle below. The given person changes to his left heel to strike at the half of this circle after first strikes his right heel, then changes back to right heel for finish. This given person, as well as all other human will perform this walking pattern in a repeatable way.

Fig.3 single walking cycle of a person

But why some people have the experience that recognizing acquainted people by the way of their walking. Because people will not only perform their walking pattern in repeatable way but also have their individual unique[9]. Psychology study showed that gender could also be distinguished due to the anatomical differences which states that women have greater shoulder swing and more hip swing. More than this, pathology study by Murray[10] shows that there appear to be 20 distinct gait components. The pelvic and thorax rotation or the inclination of the thigh[9] could also be a cue to human gait recognition, see Fig.4 below:[Automatic recognition by gait, Nixon and Carter].

Fig.4 Hip inclination measurement [Automatic recognition by gait, Nixon and Carter].

Since human gait has the uniqueness attribute, it can be developed into a new kind human biometric. Biometric is a method for uniquely recognizing humans based upon one or more intrinsic physiological characteristic, for example, fingerprint which is natural unique for every person, or some behavior traits, like hand writing which is developed into uniqueness after a long time. The biometric researches include automatic face recognition, fingerprints, voice patterns, iris identification etc. Applications based on these searches appear in our life widely from banking to security. Although with the widely use of biometric technologies, human gait as a new biometric technology is just under research in the past 10 years. But human gait as a new biometric technology has its own advantages [Automatic Gait Recognition, Nixon and cater]. The first advantage is its non-invasive extraction process. Comparing with other biometric extraction process, like fingerprint. The human gait information just tracked from a video segment with no subject contact. The second advantage is human gait less likely to be obscured because the person who under perceived need walking for movement. But other current biometric identify method tend to be easy obscured. For example, Face may be hidden by hood; palm could be protected by PVC gloves; the ear could not be seen in rather low-resolution image. However people need walking for their basic movement so that motion could always be appeared if camera can capture them. Except the perceptibility, another advantage for using human gait as biometric identification is that motion of gait is difficult to camouflage. This is especially fatal reason for using human gait for biometric for identify a suspect in crime scene. Imagine in the crime process, the criminal either does not want to focus on camouflage his walking for drawing attention or want to move quickly. Apparently, with these practical advantages that human gait has a great potential to be a new biometric method. The fourth advantages is gait can be perceived at a certain distance in a really low-resolution vedio…..

Background Subtraction

4. Current Approaches of Gait Recognition

Existing methods for human gait recognition can be divided into two main categories: Shape-based method (model-free method) and model-based method. Shape-based analysis mainly focus on recognize person from the silhouette in the sequence of images. And Model-based analysis tries to perceive the movement of the torso and the legs, which more concern on a dynamics. After 2001, Shape-based analysis divided into two aspects, one inherit and continue develop the moving shape, the other combine shape and motion. Since 2001, the Model-based analysis also evolves into two directions. Structural approaches mainly based on using statistic parameters of figure. And modeled approaches try to solve the problem based on the relative motion of the angles between limbs. Here in after, this section will review the typical approaches in all these categories.

Approx. Year

Shape-based analysis

Model-based analysis

1994 to 2000

Moving shape


Since 2001

Moving Shape

Shape + Motion


Fig.5 categories of gait recognition methods

4.1 Model-free method

Currently there is much research in analyzing human gait via computer vision techniques. From spatiotemporal patterns method by Niyogi [4] in 1994 which generates a characteristic pattern in a spatiotemporal (XYT) volume until now, the various analyses of gait techniques evolve continually. Mainly there are two branches of those analyses, one is shape-based analysis and the other is model-based analysis. Shape-based analyses focus on the silhouette of the typical human object. These kind of analyses obtain the human silhouette by derive the moving object from the image and video background. After that, the individual human object can be recognized by measurements that the person's movement or shape. Such analyses does not base on the construction of human body model but on the feature vector statistical and period of gait information in a sequence of images.

Fig.1. Shape-based analysis which extract human gait silhouette [5]

The primary of such shaped-based approaches include spatiotemporal pattern by Niyogi's XYT gait analysis, Principal Components Analysis (PCA) by Murase, Shape of motion method by J.little etc. This analysis has the advantages mainly on calculation speed and simplicity. But it has some disadvantages due to the connection is indirectly linked with gait dynamics. It is purely statistical method which based on the motion content in a long sequence of images rather than the relative human body motion. Such approaches will affect some practical effects, for example the occlusion. These approaches derived human motion within the small amount pixel in sequence images but this will be disturbed when object blocked by other object. For example, when we use shape-based approach in airport to recognize a typical person and unfortunately this person is much likely to be blocked by the luggage carrier or other people. This occlusion affection will result of removing this person temporary from the images sequence. Such affection probably direct impact the human gait extraction and finally fail. Fig.2 below show the occlusion event

Fig.2. Occlusion in human gait recognition [6]

4.1.1 Primitive model-free approaches



Shape-based analysis

Moving Shape method




Spatiotemporal pattern [1]

Shape of motion [3]

Principal Components Analysis (PCA) [2]

PCA + Canonical Analysis (CA) [4]


Sourabh A. Niyogi et al.'s spatiotemporal pattern approach generates a characteristic "braided" pattern in a spatiotemporal (XYT) volume. This approach finds the "braided" pattern and detects the contours of the walker. Also include recovery the contours of moving object in XT domain, for example, head and legs. Combining analyzing in space and time together, it used the spatiotemporal edge of the walker's body contour in spatiotemporal volume. Since the sequence of walking people contain characteristic attributes in XYT and this approach design algorithms for recognizing those attributes. But features like edges and contours' extraction from noisy images could not be extracted easily.

(First slice the volume of walker's head height, this approach use Hough transforms to find tilted stripes' parameters. Then it used a change detection operation on XT slices to distinguish the moving object, median filter for recovering background and a soft threshold on each frame and background's differences. Using the walker's head and legs to trace out )

Hiroshi Murase et al.'s Principal Components Analysis (PCA) approach improves the spatiotemporal pattern method by calculating the correlation of the spatiotemporal in the parametric eigenspace transformation (EST), simplify can be understand that it is a template matching method based on EST [free2-5]. Using such parametric eigenspace, the computational cost of correlation-based comparison between image sequences has been effectively reduced. In the end of his paper, Murase designed a lip reading experiment express the method's advantage to noise in the input images.

James J. Little et al.'s Shape of motion approach presented a concept as shape of motion, which is an instantaneous motion. Their research analysis and demonstrated the model-free feature of the walking person is variant with time but in periodic way. Also these feature variations are not only repeatable but also vary significantly between each walker. Base on that, J. Little distinguish the candidates by periodic variation in shape of their motion. In his research, he described the shape of motion with a cluster of features perceived from moments of a dense flow distribution.

P.S. Huang et al.'s PCA + Canonical Analysis (CA) approach improve PCA approach by using data analysis to increase classification capability. Huang combined canonical space transformation (CST) based on Canonical Analysis (CA) and used eigenspace transformation (EST) for character's feature acquisition. Such PCA plus CA method has better discrimination.

4.1.2 Moving shape approaches

Since 2001

Moving Shape approaches

Unwrapped silhouette [5]

Silhouette similarity [6]

Relational statistics [7]

Self-similarity [8]

Key frame analysis [9]

Frieze patterns [10]

Area [11]

Symmetry [12]

Key poses [13]


L. Wang et al.' unwrapped silhouette research described a novel gait recognition based on statistical shape analysis. A new background differences technique have been applied to extract silhouette segments from input images' background. The shape variations of these silhouettes are then described in the common coordinate system as sort of associated vector configurations. And those configurations are further analyzed using the procrustes shape analysis method to perceive eigenshape signatures which represent by the walker's shape signal. Sudeep Sarkar et al.' silhouette similarity approaches perform recognition by temporal correlation of silhouettes. Their goal has been achieved by definition of a bounding box which silhouette template matches. In their research they also stated the covariate factors that may effect on recognition of subject. Isidro Robledo Vega's relational statistics approach is a novel strategy, which differ from previous researches. Compared with other previous shape-based methods mainly focused on the attributes of individual features, this approach emphasizes the relationship of motion and feature spatial changes. This relation statistics are modeled using the probability that a random cluster of characteristics in an image have an implicitly relation. This relation statistics approach could also recognizing the running subject during the experiment.

Fig. Relational statistic of image feature

Chiraz BenAbdelkader's Self-similarity approach used PCA applied to self-similarity plots that extracted from differences in image sequences. Using the self-similarity plot (SP), this method analyzed the frequency and periodicity of the human gait. And in that way, this method could perceive the human gait. Robert T. Collins's key frame analysis based on matching subject's silhouettes from video's key frames cross a gait sequence. This method also perceived gait cues like, stride length and arm swing and moreover, the body height, width and proportion of subjects' biometric shape cues.


Yanxi Liu's Frieze patterns approach collapse the gait silhouette sequence into two-dimensional spatiotemporal patterns that combine within a time dimension. This method is similar with Niyogi's spatiotemporal pattern in XYT research, but with a specific spatiotemporal viewpoint to find the correlation and symmetry of the features' changing.

Fig. liu

4.1.3 Shape + Motion approaches

Approx. year

Shape + Motion

Since 2001

Eigen space sequences [14]

Hidden Markove Model(HMM) [15]

Average silhouette [16]

moments [17]

Ellipsoidal fits [18]

Kinematic features [19]

Gait style and content [20]

Video oscillations [21]

4.2 Model-based method

Compared with shape-based approaches, the model-based approaches model

4.2.1 Structural type approaches

Approx. Year

Model-based analysis

1994 to 2000


Single oscillator

Since 2001


Stride parameters [1]

Human parameters [2]

Joint trajectories [3]

Video oscillations [7]


4.2.2 Modeled type approaches

Approx. Year

Model-based analysis

1994 to 2000


Single oscillator

Since 2001


Articulated model [4]

Dual oscillator [5]

Linked feature trajectories [6]