Hand Gesture Database

Description

This database is composed by a set of high resolution color sequences acquired by the Senz3D sensor. It was created to validate a hand-gesture recognition system for Human-Computer Interaction. Since a mouse-like pointing device was proposed as an example of application in this work, the hand gestures in this database were designed according to mouse functionalities: cursor, left click, right click, mouse activation, and mouse deactivation. All the video sequences were recorded in a realistic scene, which means a non-uniform background, and other moving objects.

This database is composed of two sets of gestures, Set 1 and Set 2, which contain both static and dynamic hand gestures.

Set 1 (without ground-truth)

Set 1 contains 5 hand gestures performed by different people. For training purposes, there are several short video sequences per gesture. For testing, 6 long video sequences are proposed, in which 6 subjects perform different hand gestures in a continuous way as if they were using the application. This set lacks ground truth.

Set 1 is structured in 2 folders as follows:

training\

cursor\

seq_01\

seq_02\

seq_03\

...

fist\

left_click\

palm\

right_click\

test\

subject_1\

subject_2\

subject_3\

...

subject_6\

Set 2 (with ground-truth):

Set 2 contains 5 hand gestures performed by 6 subjects. For training purposes, there are 6 video sequences per gesture, each one performed by a different subject. For testing, 6 long video sequences are proposed, in which every subject performs different hand gestures in a continuous way as if he was using the application.

This set contains the ground truth of the sequences. For every video sequence in the database, there is a folder called ground_truth containing a .mat file (Matlab format) called ROIs.mat. This is a struct-type file composed of two fields:

• The first field, imageFilename, contains the file name of the images.
• The second field, objectBoundingBoxes, contains the ROI information for each image as [col row width heigth], where (col,row) is the upper left corner of the ROI.

Regarding the video sequences for testing, each ground_truth folder also contains a .txt file called temporal_segmentation.txt. It includes the beginning and ending frames of each gesture in the considered video sequence.

Set 2 is structured in 2 folders as follows:

training\

g01\

subject_1\

ground_truth\

sequence\

subject_2\

subject_3\

...

subject_6\

g02\

g03\

g04\

g05\

test\

subject_1\

grond_truth\

sequence\

subject_2\

subject_3\

...

                                        subject_6\

Download

• Set 1
• Set 2

Citation

A.I. Maqueda, C.R. del Blanco, F. Jaureguizar, N. García, “Human–computer interaction based on visual hand-gesture recognition using volumetric spatiograms of local binary patterns”, Computer Vision and Image Understanding, Special Issue on Pose & Gesture, vol. 141, pp. 126-137, Dec. 2015. (doi:10.1016/j.cviu.2015.07.009)

For questions about this dataset, please contact This email address is being protected from spambots. You need JavaScript enabled to view it. or This email address is being protected from spambots. You need JavaScript enabled to view it..