Analysis Of Partial Occlusion On Human Pedestrian Detection

Thevendran, Sasinthiran (2018) Analysis Of Partial Occlusion On Human Pedestrian Detection. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (24 Pages) Analysis Of Partial Occlusion On Human Pedestrian Detection.pdf - Submitted Version Download (800kB)

Abstract

Over the last few years, recognizing human in pedestrian surveillance system video is crucial for diverse application area.For an example,it can overcome the terrorism,some general social problems and violence by providing surveillance to the pedestrians.This also ensures to keep them in a very close watch for a secured environment.However,the main difficulty in human recognition is the occlusion of human and the light intensity.The common problems that influence the performance of human recognition system is the light intensity,distance of the camera,colour of the shirt or different poses of human.Occlusion can be happen in different ways,human with human or human with objects.In this system, a Webcam camera is used to analyse the recognition when the occlusion happens.Python software, OpenCV library and Keras API are used to train the system to recognize the human.Haar Cascade is used to track humans and the Convolutional Neural Network is used to classify the human models.The webcam camera captures the image and recognizes the human that passes by it.The data is taken from the real time video of the system and is used to analyse the accuracy and performance of the system.The analysis proved that this system has high accuracy in recognizing human which is 97.35% of accuracy.It only has 2.65% rate of error.The occlusion analysis shows that the accuracy decreases as the occlusion increases.Then,the results clearly shows that moderate lighting is the perfect lighting for this system and when the distance increases the accuracy decreases. The distance problem maybe caused by the ability of the camera.So,this system performs very well in human recognition and it also has high accuracy.

Actions (login required)

View Item