Backgroung Subtraction Based Detection and Tracking Of People In Video
Shashank Joshi, J. R. Panchal · 2014
People Detection in video is generally used in high level multimedia applications like intelligent surveillance systems, augmented reality ,etc. People detection is based on background subtraction. Generally videos are available in compressed form due to which noise will be added in it. Many algorithms are used to detect people and control their rate. There are three key steps in video analysis detection of interesting moving objects, tracking of such objects from frame to frame, and analysis of object tracks to recognize their behavior. The ability of human visual system to detect visual saliency is extraordinarily fast and reliable. However, computational modeling of this basic intelligent behavior still remains a challenge. In this paper we put our attention into background subtraction and Gaussian grouping of pixels for detection of people in low quality video i.e. improve accuracy. I. INTRODUCTION The field of computer vision is concerned with problems that involve inter- facing computers with the environment through visual means. The increase of high-powered computers and the availability of high-quality and inexpensive video cameras extend the computer vision's applications to everyday life technology. People tracking is one of the most important tasks in computer vision which can be defined simply as the problem of estimating the trajectory of each person in the image plane as s/he moves around in the scene. In other words, a people tracker system can recognize each person in consecutive frames of a video. Depending on the applications of people tracking, additional information can be also provided by the system during tracking. Background subtraction technique find the foreground object from video and then classify it into categories like human, animal, vehicle etc., based on shape , color , motion or other features. Most of the multimedia videos are available in compressed format. Usually higher the compression rate, lowers the correct hits and quality of video due to noise added in it. A modern object detection algorithm can be divided into five parts: pre-processing and normalization, local rectification and compensation of small shape variations, computation of descriptor set, machine learning classification, and post-processing to fuse multiple detections. . In this paper, we focus on detection schemes based on background subtraction because of their widespread use and the possibilities they offer in implementing real-time object detection systems. background subtraction is nothing but foreground detection. Any motion detection system based on background subtraction needs to handle a number of critical situations such as, noise image due to a poor quality image source, variations in the lighting conditions, small movement of non-static objects, shadow regions that are projected by foreground objects, multiple objects moving in the scene both for long and short periods. The main objective of this paper is to develop an algorithm that can detect people. Unfortunately, images and video are usually available in compressed format which makes object detection more difficult because of the additional distortion noise. In this paper we propose a saliency map algorithm and compare it with background method to improve accuracy. The process algorithm is described as follows: 1. Sequences of Video Frames