Action recognition

@author: Junhao Hua

Action Recognition & Categories via Spatial-Temporal Features

This project is mainly based on Prior Dollar's work:

Dollár, Piotr, et al. "Behavior recognition via sparse spatio-temporal features." Visual Surveillance and Performance Evaluation of Tracking and Surveillance, 2005. 2nd Joint IEEE International Workshop on. IEEE, 2005.

toolbox

http://vision.ucsd.edu/~pdollar/toolbox/doc/index.html

Runing MyDemo.m

Feature Extraction

1. set is_Feature_extract = 1, remove comment

   conv_movies2clips featuresLGdetect featuresLGpca featuresLGdesc

   get and save DATASETSprLG.mat

Classificaton

2. If you just want to test a video sample using KNN or SVM, try

(1) set is_SingleAction_train = 1

(2) then, is_SingleAction_Recog = 1

a) knn b) or svm

3. If you want to test all video samples

(1) using knn, set is_Action_knn_Classif = 1

(2)or using svm, set is_Action_svm_Classif = 1

4. if you want to use voting methods

(1) if just test a video sample, set isCreateCodeBook = 1 set is_vote_Action_Recog = 1

(2) if test all sample, set is_vote_recog_test = 1

DATASET format

Some DATASETS can be downloaded from Baidu Cloud (数据集百度云下载).

Weizmann human action dataset

(download from : http://www.wisdom.weizmann.ac.il/~vision/SpaceTimeActions.html )

set00/

ira_walk.avi

moshe_jump.avi

moshe_run.avi

...
after running feature extraction, following data generated

clip_ira_walk.mat

cuboids_ira_walk.mat

features_ira_walk.mat

...

set01/

set02/

...

set08/

file directory setting

TestDir.m
OutputDir.m
datadir.m