Identify level of interest of people to a object using kinect

Abstract

This paper explores whether it is possible to detect peoples interest toward an object by using a kinect camera. The kinect camera has a depth and a color camera and the most interesting part to use is the depth camera..The depth camera is interesting due to that each pixel contain the distance from the camera to the object. With a normal color camera each pixel contain the color value. The main focus in this master thesis have been to develop a system that can detect whether multiple people are looking toward an object by using kinect camera. The system consists of a kinect SDK and a Mac computer running the program. The paper starts by giving an introduction to the software-tools used and give the specifications of the hardware. Blob detection and face detection is the two computer vision methods used. The basic contents behind the methods are described in the following chapters. Blob detection is used on the frames from the depth camera. Face recognition is used on the frames from the color camera. In this paper the openFrameworks library is used. The most important library from openFrameworks is the OpenCV and kinect library. The OpenCV library has algorithms for face detection and blob detection. How these two methods are integrated to the system is described in the report. The main challenge has been to implement the system and integrate the two methods. The system created is able to detect objects contour using the depth image. Then do a face detection on the detected object. This is to determine if the detected object is a person looking towards the camera. The system is tested on a test described in this paper. This test is designed to find out if the system is able to do what it's intended to on a simple scenario and more advanced scenarios. The test is also designed to find the limitations of the system. The results from the test is documented in the report and discussed.