International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395 -0056
Volume: 04 Issue: 05 | May -2017
p-ISSN: 2395-0072
www.irjet.net
Study of Different Action Recognition Techniques Nikita Pachpute1, Priya Kamate2, Gayatri Walmik3, Shivani Jadhav4 IT Department, MIT College of Engineering, Pune , Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In this paper we propose three approaches for
human action recognition from input video stream:1) action recognition from silhouette images and 2) action recognition by using hidden markov model(HMM). 3) action recognition by using depth map resources (HDMM). First approach makes use of spatio-temporal body parts movement (STBPM) for extracting spatial and temporal features from silhouette image and RAC classifier for classifying human action based on extracted features. Second approach makes use of hidden markov model (HMM) which represents process as a set of states. Third approach use rotation and class score fusion.
Key Words: Silhouette image, STBPM, RAC, HMM,HDMM, Activity recognition
1.INTRODUCTION Now-a-days Human action recognition plays very important role in various applications such as human–compute interaction, human activities analysis, and real time surveillance systems. In daily life human performs different activities which are composed of several subtasks. These day to day activities include walking, running, dancing etc. These activities may be performed at different times and may have chronological relationship with each other. Motive for writing this paper is to recognize these activities and analyze human behavior. Recognizing human action is now becoming the topic of interest in the field of video surveillance. This paper describes two different approaches for action recognition. The contribution of this work consists of two parts. Firstly we extract the features of skeleton and then we classify the activity based on those features. Three main complexity issues, as mentioned in [1], are generally present in any Human Action Recognition technique are - i) Environmental complexity: The process of Human Action Recognition depends on the quality of the video, which differs due to the environmental condition of the scene elements and makes the procedure more complex. This type of complexity includes occlusions, clutter, interaction among multiple objects, changing of illuminations etc. ii) Acquisition complexity: Besides environmental condition, the quality of the video also depends on video acquisition, which varies with respect to view point, movement of the camera etc. iii) Human action complexity: In general sense, human actions are of varied in nature, hence exact determination of human action is a complicated task. Presence of multiple human entities makes any Human Action Recognition technique more complex. © 2017, IRJET
|
Impact Factor value: 5.181
|
To handle those three issues mentioned above, some constrains have been made in [2].this constraints are like as i) use the videos, which contain human silhouettes only and didn’t consider any silhouette extraction techniques also. Video So, the environmental and acquisition complicacies are avoided. ii) To reduce the complexity, the proposed work considers videos containing only one human object in each of the frame. iii)consider that the head should be in the upper portion of the body and the body should not be upside down. 2.RELATED WORK Computer-vision-based human motion analysis has become an active research area. It is strongly driven by many promising applications such as smart surveillance, virtual reality, advanced user interface, etc. Recent technical developments have strongly demonstrated that visual systems can successfully deal with complex human movements. Human Action Recognition task mainly classified in three phases according to [5] as follows i)motion segmentation ii)tracking iii) Behavior understanding. In motion segmentation the image is divided into regions as foreground and background. Foreground image should be an object such as person, cars, animal etc Motion segmentation separates foreground images from background images. Object tracking is comes after the motion segmentation. Tracking is a particularly important issue in human motion analysis since it serves as a means to prepare data for pose estimation and action recognition. By using Tracking we can estimate the pose of the object and recognize the action of object. This can be done on the basis of points, lines, blobs etc. After the tracking we go for the Behavior understanding. Behavior understanding contains action recognition and description. 3.ACTION RECOGNITION FROM SILHOUETTE IMAGES 1) Silhouette Image: Silhouette image is the image of a person, animal, object or scene usually represented as a solid shape of single color. Silhouette image has only two color levels and it provides shape based information. Interior of silhouette image is featureless.
ISO 9001:2008 Certified Journal
|
Page 827