Detection and Classification of a Moving Object in a Video Stream

Abstract

In this paper we present a new method for detecting and classifying moving objects into humans and vehicles from a video surveillance scene. In this approach, the moving objects are firstly detected from the background using a background subtraction technique. Background subtraction algorithms are implemented in a MATLAB environment. A comparison for all the algorithms was made to determine which background subtraction algorithm performs better with the proposed classifying algorithm. The algorithms were then tested using more than a video and many frames of the video was tested. Secondly, edge detection of moving objects was performed using Canny or Prewitt operations, while bounding boxes were implemented over the moving objects. Edge detection and bounding boxes were used before the classification step to simplify the classification. Finally, classifying the moving object into humans and vehicles was accomplished by finding the height-width ratio of the bounding box around the moving object in each frame and estimating the speed of the moving object from two consecutive frames in the video stream. The results found were fairly good and the wrong classification was due to the bounding boxes not correctly covering the moving objects. The background subtraction step and the classification step were tested using different video sequences. A comparison was made between the combination of each background subtraction technique and the classification algorithm; the false classification results are given. Although the proposed classification algorithm in this paper is simple and easy to be implemented, the results obtained are very satisfactory and the accuracy is very good