An Effective Dynamic Gesture Recognition System Based on the Feature Vector Reduction for SURF and LCSPABLO BARROS, NESTOR JÚNIOR, JUVENAL BISNETO, BRUNO FERNANDES, BYRON BEZERRA, SÉRGIO FERNANDES.

ESCOLA POLITÉCNICA DE PERNAMBUCO - UNIVERSIDADE DE PERNAMBUCO - BRASIL

RPPDI Dynamic Gesture Recognition Database

Dynamic Gesture Frame Sequences

Represent one Gesture

http://rppdi.ecomp.poli.br/gesture/database/

Dynamic Gesture Recognition System

System Architecture Feature Extraction Module

Classification Module

Extraction Module

Local Contour Sequence – LCS [1]

Speed Upt Robust Features – SURF [2]

Convexity Approach CLCS

CSURF

Local Contour Sequence - LCS

Algorithm Identify Hand Shape

Image Segmentation

Contour Detection

Calculate Feature Vector

LCS – Segmentation

Segmentation OTSU [3]

Contour Identification

Hand Contour Identification

LCS - Local Contour Sequence

Feature Vector Calculation Find the top first point of the image

Order the points in clockwise.

Calculate distance of a line formed by two points , ℎ[ −( −1)⁄2] and ℎ[ −( +1)⁄2] , and h[i].𝑖 𝑤 𝑖 𝑤

Speed Up Robust Features - SURF

Integral Image

Find Interest Points

Describe Interest Points Intensity

Direction

Descriptors

Convexity Approach

Minimize the hand shape Douglas-Peucker Algorithm

Apply convex hull Sklankys Algorithm

Calculate points distances

Convexity Approach

Douglas Peucker Algorithm Select the two most distant points.

Verify if there is vertex near than a distance T, if there is, remove it.

Recursively do it again with all the points.

Convexity Approach

Sklanky´s Algorithm Find a convex vertex.

Rename the other vertex in clockwise, starting with p0.

If p0, p1 and p2 turn right:

Put p0 after p2.

Update p0, p1 and p2.

Else:

Put p1 before p0.

Remove p1.

Update p0, p1 and p2.

Repeat until p0 is the initial vertex and p0, p1 and p2 turns right.

For each pair of points draw a line and find the most distant point.

Convexity Local Contour Sequence

Calculate distance Adaptation of LCS

Use the two external points to draw the line.

Use the inner point to calculate distance.

(a) LCS. (b) SuRF interest points. (c) CLCS. (d) CSURF

Classification Module

Elman Recurrent Neural Network

Hidden Markov Model

Dynamic Time Warping

Results

Convexity Approach

Methodology Run 30 times

Validation (1/3 for test and 2/3 for training)

Referências

[1] Meena, S. 2011. A Study on Hand Gesture Recognition Technique. Master’s thesis, National Institute Of Technology, Rourkela,India

[2] Bay, H., Ess, A., Tuytelaars, T., Van Gool, L.: Speeded-up robust features (surf). Comput. Vis. Image Underst. 110(3), 346–359 (Jun 2008),http://dx.doi.org/10.1016/j.cviu.2007.09.014

[3] Bao, J.; Song, A.; Guo, Y.; and Tang, H. 2011. Dynamic hand gesture recognition based on surf tracking. In Electric Information and Control Engineering (ICEICE), 2011 International Conference on, 338 –341.

[4] N. Otsu. A threshold selection method from gray-level histograms. Systems, Man and Cybernetics, IEEE Transactions on, 9(1):62 –66, jan. 1979.