A Capstone project report submitted in partial fulfillment of the requirements for the Degree of Undergraduate of Area Name in Mechatronic Engineering, School of Engineering and Technology

Machine Vision is a technology which involves detecting and recognizing of objects in a video or an image. Human eyes are sensitive for electromagnetic waves ranging from 390-770 nms. Video cameras can be sensitive to a range much wider than a human eye. This study is focusing on implementing a method for the machines to see through objects (in this case, humans) as a human eye. For this implementation, the detection speed and the accuracy of the detection are two main parameters that should be considered. Thus an implementation of a robust human detection method using open source computer vision (OpenCV) libraries was done. To achieve the wide range of detection space, an IP camera with PTZ controlling system is used. For a robust human detecting method, HOG with linear SVM has been selected. In order to increase the computational performances and achieve a real time detection, AMD OpenCL parallel programming was used. The system accuracy was determined under different conditions of pan, tilt and zoomed. The accuracies ranged between 97-99% in pan and tilt operations and 76% accuracy in zooming were achieved. Detection speed was enhanced by four times by the conventional HOG descriptor as well. This system can be further improved and enhanced for better use of human detection for the benefit of the technology of Machine Vision.

Capstone Project (B. Sc.) - Asian Institute of Technology, 2014