Joo Kooi Tan

Motion history image: its variants and applications

The motion history image (MHI) approach is a view-based temporal template method which is simple but robust in representing movements and is widely employed by various research groups for action recognition, motion analysis and other related applications. In this paper, we provide an overview of MHI-based human motion recognition techniques and applications. Since the inception of the MHI template for motion representation, various approaches have been adopted to improve this basic MHI technique. We present all important variants of the MHI method. This paper points some areas for further research based on the MHI method and its variants.

Human activity recognition: Various paradigms

Action and activity representation and recognition are very demanding research area in computer vision and man-machine interaction. Though plenty of researches have been done in this arena, the field is still immature. Over the last decades, extensive research methodologies have been developed on human activity analysis and recognition for various applications. This paper overviews various recent methods for human activity recognition with analysis. We attempt to sum up the various methods related to human motion representation and recognition. We make an effort to categorize the recent methods from the best in the business, and finally figure out the short-comings and challenges to dig out in future to develop robust action recognition approaches. This work exclusively endeavors to encompass the researches related only to human action recognition mainly from 2001 till-to-date with critical assessment of the methods. We also present our work along with to solve some of the shortcomings. It will widely benefit the researchers to understand and compare the related advancements in this area.

High-Speed Human Motion Recognition Based on a Motion History Image and an Eigenspace

This paper proposes an efficient technique for human motion recognition based on motion history images and an eigenspace technique. In recent years, human motion recognition has become one of the most popular research fields. It is expected to be applied in a security system, man-machine communication, and so on. In the proposed technique, we use two feature images and the eigenspace technique to realize high-speed recognition. An experiment was performed on recognizing six human motions and the results showed satisfactory performance of the technique.

Tracking of Moving Objects by Using a Low Resolution Image

Automatic detection and tracking of moving object is very important task for security system, monitoring activity and surveillance application. It is, however, still under the developmental stage and needs to be robust performance when applied in an unconstrained environment. Many approaches have been developed to detect motion, namely, optical flow, frame difference, background subtraction and skin color extraction. These methods are sensitive to illumination changes and small movement in the background such as moving leaves of trees that will cause inaccurate detection. Some techniques have been applied to reduce this kind of noise. In this paper we propose a new method by using a low resolution image to reduce that kind of noise so we can get an accurate tracking object.

Analysis of Motion Self-Occlusion Problem Due to Motion Overwriting for Human Activity Recognition

Various recognition methodologies address to recognize and understand varieties of human activities. However, motion self-occlusion due to motion overlapping in the same region is a daunting task to solve. Various motion-recognition methods either bypass this problem or solve this problem in complex manner. Appearance-based template matching paradigms are simpler and hence these approaches faster for activity analysis. In this paper, we concentrate on motion self- occlusion problem due to motion overlapping in various complex activities for recognition. In the Motion History Image (MHI) method, the self-occlusion is evident and it should be solved. Therefore, this paper compares our directional motion history image concept with basic the Motion History Image, Multi-level Motion History representation and Hierarchical Motion History Histogram representation to solve the self-occlusion problem of basic the Motion History Image representation. We employ some complex aerobics and find the robustness of our method compared to other methods for this self-occlusion problem. We employ seven higher order Hu moments to compute the feature vector for each activity. Afterwards, k-nearest neighbor method is utilized for classification with leave-one-out paradigm. The comparative results clearly demonstrate the superiority of our method than other recent approaches. We also present several experiments to demonstrate the performance and strength of the DMHI method in recognizing various complex actions. Index Terms—MHI, DMHI, MMHI, HMHH, motion recognition, feature vector

Deformable Shape Recovery by Factorization Based on a Spatiotemporal Measurement Matrix

An efficient technique based on factorization is presented for recovering the 3-D shape of deformable objects. The original factorization solely recovers the 3-D shape of rigid objects. The present technique applies it to shape recovery of deformable objects. In the proposed technique, multiple video cameras at fixed locations take video images of an object. The obtained image sequences are analyzed to yield measurement matrices at respective sample times and they are merged into a single matrix called a spatiotemporal measurement matrix. Factorization is applied to the matrix once and this results in the entire 3-D shape recovery during observation in a single stage. In the performed experiment, some human motions and the inflation/deflation process of a toy balloon successfully recovered their 3-D shape. Advantages of the present technique are that calibration is not necessary with the employed multiple video cameras, that the entire deformation is recovered by single application of the factorization, which eliminates the alignment of recovered shape in the time axis, that the recovered shape contains less recovery errors because of averaging effect by the employment of the proposed matrix, and that the technique is applicable to any object on which feature points can be specified. Various applications related to human motions recovery are expected.

Motion recognition approach to solve overwriting in complex actions

Motion overwriting due to motion self-occlusion is a big concern in motion and activity recognition. This paper presents a directional motion recognition approach that can solve the motion overwriting for complex actions or activities. Optical flow is split into four directions to compute motion templates. These templates are used to create feature vectors by Hu moments. Very satisfactory recognition results are achieved for various complex actions, which encompass motion overwriting. This method is compared with the basic motion history image method and multi-level motion history image method. The latter method professed that it can overcome motion self-occlusion problem and hence we compare these methods for several complex datasets with complex dimensions.

High Accuracy and Real Time Recognition of Human Activities

Automatic recognition of human motions or activities employing a camera and computer system has been one of the main topics in computer vision. In this paper, we propose a method of recognizing human activities/motions employing multiple cameras that surround a human in motion. For the recognition, we employ JK Motion database and an eigenspace method. A motion/activity obtained from a camera is expressed by a point. M motions obtained from P cameras are therefore described by MtimesP points in the eigenspace. Comparison of the proposed technique with other techniques is done. Experimental results of applying our approach to six motions with four observation cameras in the real world are given to demonstrate the effectiveness of the proposed technique.

High-speed Data Retrieval in an Eigenspace Employing a B-tree Structure

In this paper, we propose a technique for retrieving a database containing a large amount of data at high-speed for the purpose of recognizing human postures/motions. There is a technique for recognizing a 3D object from its 2D image set. But it has a disadvantage that the retrieval time increases in proportion to the number of objects registered in the database. We have developed a high-speed retrieval technique of such a large database by applying B-tree data structure to the database. An experiment was performed employing some human posture images as well as alphabetical images and the proposed technique achieved satisfactory results

Automatic Detection of Lung Nodules in Temporal Subtraction Image by Use of Shape and Density Features

Computer aided diagnosis is one of the most important tools for supporting system on visual screening in medical fields. However, sensitivity for detection of small nodules is unsatisfactory. It is because detection of subtle lesions on computed tomography (CT) images is a difficult task for radiologists. Recently as one of solutions to avoid the problem, a temporal subtraction technique is introduced. The temporal subtraction image is obtained by subtraction of a previous image from a current one, and can be used for enhancing interval changes on medical images by removing most of the normal background structures. In this study, we have developed a new method for automatic detection of lung nodules based on artificial neural networks from a temporal subtraction image. First, the candidates for nodules were detected by use of a multiple threshold technique based on the pixel value in the temporal subtraction images obtained by the voxel-matching technique. Next, false positives of nodules candidate are removed by use of selective enhancement filter and a rule-based method with classifier based on artificial neural networks. We applied our computerized scheme to 6 MDCT cases including 87 lung nodules. Our scheme for detecting lung nodules provided a sensitivity of 80.5% for lung nodules with sizes less than 20mm, and with 7.5 false positives per scan. In this paper, we discussed the experimental result of detection and statistical features.