# Human-Activity-Recognition-from-Videos
Nowadays, it’s a very hot topic on video-based human action detection, which has recently been demonstrated to be very useful in a wide range of applications including video surveillance, tele-monitoring of patients and senior people, medical diagnosis and training, video content analysis and search, and intelligent human computer interaction [1]. As video camera sensors become less expensive, this approach is increasingly attractive since it is low cost and can be adapted to different video scenarios.
Actions can be characterized by spatiotemporal patterns. Similar to the object detection, action detection finds the reoccurrences of such spatiotemporal patterns through pattern matching. Compared with human motion capture, which requires recovering the full pose and motion of the human body, the task of action detection only requires detecting the occurrences of a certain type of actions. Video features for action detection The development of video-based action detection technology has been ongoing for decades. The extraction of appropriate features is critical to action detection. Ideally, visual features are able to handle the following challenges for robust action detection:
1. Viewpoint variations of the camera
2. Performing speed variations for different people
3. Different anthropometry of the performers and their movement style variations
4. Cluttered and moving backgrounds.
Previously, human bodies were tracked and segmented from the videos to characterize actions and motion trajectories are popularly used to represent and recognize actions. Unfortunately, only limited success has been achieved because robust object tracking is itself a nontrivial task. Recently, interest point based video features show promising results in the action detection research. Such interest point-based video features do not require foreground/background separation or human tracking [2]. We searched a lot and found many techniques to identify actions in videos, like space-time interest point (STIP), which is developed by Laptev and Lindeberg. STIP features have been frequently used for action recognition. However, the detected interest points are usually quite sparse, and it is time consuming to extract STIP features for high-resolution videos. And then we finalized to work on few types of interest-point based feature extractions like;
1. The first type of interest point features is called 3-D SIFT, developed by Scovanner et al [3]. This descriptor is similar to scale invariant feature transformation (SIFT) descriptor except that the gradient direction for each pixel is a three-dimensional vector. It can work with any interest point detector.
2. The second type of interest point features is named spatiotemporal interest point (STIP) [2].
3. The third type of classification is done by using Histograms of Oriented Optical Flow (HOOF), Histogram of Oriented Optical Flow (HOOF) features are independent of the scale of the moving person as well as the direction of motion. Extraction of HOOF features does not require any prior human segmentation or background subtraction.
没有合适的资源?快使用搜索试试~ 我知道了~
资源推荐
资源详情
资源评论
收起资源包目录
Human-Activity-Recognition-from-Videos-master.zip (72个子文件)
Human-Activity-Recognition-from-Videos-master
README.txt 2KB
Human Activity Recognition.pptx 734KB
testVideos
person01_walking_d1_uncomp.avi 2.28MB
person01_jogging_d3_uncomp.avi 1.63MB
report_HumanActivityDetction_SSI_GopikrishnaErabati_MohitKumarAhuja.pdf 847KB
MatlabCode
actionDetection.m 15KB
FeaturesTrainSTIP
testFeatLabelsKTHNew.mat 213B
trainFeaturesKTHNew.mat 2.61MB
trainFeatLabelsKTHNew.mat 380B
trainFeaturesKTHNew1.mat 773KB
trainFeatLabelsKTHNew1.mat 326B
testFeatures.mat 15KB
natsortfiles.m 5KB
STIP
padSame.m 635B
natsortfiles.m 5KB
dir2.m 390B
maxSupression.m 1KB
computeSTIP.m 3KB
trainData.m 2KB
natsort.m 12KB
crop2.m 211B
pad.m 262B
recognizeAction.m 546B
myGaussFft.m 392B
testData.m 1KB
getAccuracy3DSift.m 3KB
dir2.m 390B
3DSIFT
buildOriHists.m 2KB
AddSample.m 445B
PlaceInIndex.m 410B
KeySampleVec.m 715B
README_3DSIFT.txt 2KB
KeySample.m 2KB
sphere_project.m 2KB
LoadParams.m 809B
MakeKeypoint.m 750B
sphere_tri.m 6KB
NormalizeVec.m 108B
GetGradOri_vector.m 1KB
mesh_refine_tri4.m 4KB
Create_Descriptor.m 1KB
natsort.m 12KB
3DSift_ourfiles
trainData3DSIFT.m 3KB
convertVideoToMat.m 1KB
natsortfiles.m 5KB
getAccuracy3DSift.m 3KB
dir2.m 390B
testData3DSift.m 3KB
natsort.m 12KB
buildClassLabel.m 540B
featureSet3DSift.m 782B
interestPoints.m 1KB
FeaturesTrainOpticalFlow
testFeatLabelsKTHNew.mat 213B
trainFeaturesHOOF.mat 1.82MB
trainFeatLabelsHOOF.mat 246B
actionDetection.fig 24KB
getAccuracyHOOF.m 3KB
FeaturesTrain3DSift
testFeatures.mat 2.89MB
trainFeatures.mat 5.32MB
getAccuracySTIP.m 3KB
OpticalFlow
natsortfiles.m 5KB
LucasKanadeRefined.m 3KB
dir2.m 390B
natsort.m 12KB
trainDataHOOF.m 2KB
LucasKanade.m 2KB
gradientHistogram.m 2KB
HierarchicalLK.m 3KB
recognizeAction.m 476B
testData.m 1KB
Reduce.m 554B
README.md 3KB
共 72 条
- 1
资源评论
JaniceLu
- 粉丝: 78
- 资源: 1万+
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功