Kinect Based Human Identification Method Using Machine Learning Techniques

In this work, we address a problem of people identification. People identification is an important feature in various applications like using in banks, airports, border crossings etc. For purpose of people identification today are used different methods such as face recognition, fingerprint, scannin

Project Title

Kinect Based Human Identification Method Using Machine Learning Techniques

Project Area of Specialization

Computer Science

Project Summary

In this work, we address a problem of people identification. People identification is an important feature in various applications like using in banks, airports, border crossings etc. For purpose of people identification today are used different methods such as face recognition, fingerprint, scanning of eye retina, voice recognition etc. The most of these methods require interaction with people while one method, people gait recognition, can be proceeding even without awareness of people who is in process of identification. Because of that, people gait recognition is interesting field in identification process and biometrical techniques.

Gait recognition is an important type of biometric that aims to recognize human gait based on their walking style. This paper takes advantage of the Microsoft Kinect sensor to get in-depth information on the human skeleton and trajectory of joints for recognition. In this paper, we investigate two sets of features which were extracted during one gait cycle. The first set is called Static Features (SF), which is based on the length of bones (upper and lower legs, Torso, upper and lower arm). The second set is Dynamic Features (DF), which were extracted from the height to the ground of (wrist, shoulder, and ankle) and distance between shoulders during one gait cycle. Extracting these features is difficult when using traditional camera such as CCTV. A database has been created to perform our experiments; this database consists of twenty participants walking, when indoors, form right to left. For each person 10 videos have been recorded. We used K-nearest neighbor (KNN) as a classification method, based on Cityblock distance. The experimental result of the proposed method accomplished a success rate of 95.5%. The experimental result also shows that the proposed provides a significant result compared to the existing methods.

Project Objectives

Our main objective is to deal with current issues faced in the identification of a human being. We plan to make this process more accurate, less hectic and more easily applicable in all circumstances without any limitations and difficulties.

Project Implementation Method

Agile methodology is being used in our project development. In agile methodology, our team divided the project to several stages in which they will involve constant collaboration and continuous improvement and iteration at every stage Database Design.

Benefits of the Project

1. Direct Customer/Beneficiaries of the Project

Direct customers are those who will contract with our team with respect to the business or purchases products comprised within the Business from the Seller’s Group.Direct users in our project can be a commercial business company which requires Human Identification in any way or the person holding this project.

2. Outputs Expected from the Project

Our project will gave some outputs as follow:

i) It will identify the specific human even without awareness of people who is in process of identification.

ii) It will profit our stakeholders in the form on Security.

Technical Details of Final Deliverable

We intend to do Human Identification using a Microsoft Kinect sensor device. This is a device which was initially made for gaming purposes, and came with the Microsoft Xbox gaming console. Though, lately, it has been used for other purposes as well. This device detects human presence within its range of 0.5 to 4.5 meters, and draws a virtual skeletal model of the subject. The device neither needs the subject to face it, nor does it require any light to detect. It works accurately in the dark as well. 20 different joints are detected and their respective positions are extracted. An example of such a model has been attached below in the form of an image. These 20 positions are fed to the system by the device. We intend to use this data to generate further data, such as the bone length of the subject, the limb length, size of the upper body, size of the lower body, the ratio of two bones of the subject, the shoulder width of the subject etc. 

In the first phase of our project we will collect data and store it in a database. After the collection of enough data to train our model, we will try to predict the test cases. We intend to use multiple techniques to gain the optimum accuracy. This may include the classification of the collected data into various classes, for example, on the basis of heights, or widths, or bone lengths etc. Using classification, we can assign the test subject to a specific group and then match the data of the subject in that specific class. We can then either predict, who the subject is, explicitly, or assign different probabilities of him/her being a specific person.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Security

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Industry, Innovation and Infrastructure

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com