Font Size: a A A

Gait Recognition Algorithm Based On Extraction Of Joints And Multi-View

Posted on:2009-06-25Degree:MasterType:Thesis
Country:ChinaCandidate:L XiangFull Text:PDF
GTID:2178360245955309Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
Gait recognition is a new technology of biometric authentication which absorbs more and more researchers to concern it in recent years; it recognises people by theirway of walking. As a bio-authentication techniques, gait recognition have the unique advantages which other biological authentication technology does not have, such as recognize people far away, in low video quality, also, it is not easy to hide gait.Gait analysis aims at attempting to detect, track and identify people, and more generally, to understand human behaviors. This paper pays more attention on gait detection, especially on the preprocessing of gait, feature extraction and classification. We also generalize the research state and used methods in and out.First, the methods of image preprocessing is studied. Obtaining the silhouette of image is the direct processing of background difference images. Then, a comparison on image threshold segmentation algorithm, including iteration method, double-peak histogram conventional algorithm, otsu method and statistical method are taken. These get an excellent goal. Next, the opening operation, close operation, morphological dilation and erosion needed are taken to eliminate the noise and to obtain smooth and the connected human body image.Next, research of gait recognition method based on side view is taken.Through studies the human body to walk the sample periodicity to gain the key frames that represents the human body periodic gait characteristic, carrying on the preprocessing to the essential frame, body silhouette is segmented from the image and that is converted into a skeletal model afterwards. Furthermore joints coordinates is got and angle characteristic of joints is got, quantity of process the characteristic relatively is quite big, therefore analyzes (PCA) with the traditional application principal components to carry on the feature extraction and the compression. After feature space many step transformations the obtained weight vector distinguishes. At last, ENN (Nearest-Neighbor with class Exemplar) and leave-one-out cross validation method for classification and recognition. From the database which constructed 6 person sample 2 periodic sequences personally, the side view (is opposite in background 0 degrees) after all of these gait recognition method of skeletal model has made certain progress.Then, study is carried of two kinds of gait recognition algorithms under the multi-view based on the human body skeleton model.Implementation of reflection and emission algorithm based on multi-view on is taken. The original algorithm is only the human body actual spot object picture element mapping, the author introduces the transformation operator the body length relates the human body limb virtual length and in the corresponding image. The transformation operator definition is the actual body length and the corresponding ratio of sketch body length, through seeking human body sample straight line walk in angle and sketch information and so on body length can obtain this kind of operator, the body length substitutes the sketch in this transformation operator to obtain again the corresponding actual human body limb length.Moreover, research of curve fitting algorithm based on the multi-view is taken. The human body image both feet center point y-coordinate is the independent variable, the human body actual body length and with its corresponding ratio of image body length is a dependent variable. Height of the human body sample is known, these are to construct from constructs under the database the curve. The author has compared 10 kinds of curves, obtains the best curve, estimates the complete sample through the best curve this kind of mapping relations.Two kinds of multi-view algorithms to the own establishment's 6 samples, 2 sequences, 3 directions (is opposite in background 30 degrees, 45 degrees, 60 degrees) under and use KNN and leave-one-out cross validation method for classified recognition, and has made the good progress. Two algorithms about has been suitable the condition, the limiting condition, the performance characteristic comparison are shown.
Keywords/Search Tags:Gait Recognition, Skeletal Model, Multi-View, Reflection and Emission Method, Curve Fitting, Principal Components Analysis
PDF Full Text Request
Related items