A Review on Moving Object Detection

Abstract: Video surveillance is a hot debated topic around the world for its various applications in different important fields like security, traffic camera, bank, museum, educational institutions and many other fields. The main issue behind this “video surveillance” is the movement of object. Moving object detection refers to the idea of identifying the physical movement of an object in a given area. For moving object detection, the current frame and the background frame of a video sequence is under consideration. This paper attempts to give a review on the ongoing and past remarkable and mentionable works on moving object detection.

Introduction

Video is an electronic medium for the recording, copying, playback, broadcasting, and display of moving visual media. Surveillance refers to the close observation of a person or object with the intention of influencing, managing, directing and protecting. Video surveillance is an appliance that enables embedded image capture capabilities that allows video images or extracted information to be compressed, stored or transmitted over communication networks or digital data link. A video is basically a group of frames where frame can be defined as one of many still images that combine into the total moving picture. Moving object detection is the concept of differentiating the stationary objects from the moving or non-stationary objects with respect to a given region or area. There are basically 3 steps into video analysis namely, identifying, tracking and analysis. And as we can see that is not possible without moving object detection. Moving object detection has turned into a very serious topic for the researchers, scientists and in fact anyone who works with video surveillance and analysis. The main reason behind that is the application of moving object detection in various vast fields.

We can name some of the fields as video surveillance, monitoring of security at airport, law enforcement, video compression, automatic target identification, marine surveillance, radar system, theft prevention, departmental stores and many other areas.

There are a lot of techniques used for moving object detection but a few like background subtraction, frame differencing, temporal differencing and optical flow can be specially mentioned due to their extensive use as traditional methods. But each of these techniques has disadvantages associated with them.

Although new techniques are invented and applied every day but still that is a challenging task due to dynamic background, illumination variations, misclassification of shadows as object, camouflage and bootstrapping problems. Many researches are carried out with respect to these problems which this paper attempts to shade a light on.

The remaining of this paper is indexed as follows:

Section II. Usual traditional approaches

Section III. Recent Researches

Conclusion

Usual traditional approaches

The figure below will try to classify and categorize the moving object detection approaches.

A. Frame Differencing

Frame differencing, as the name suggests, detects moving object by differentiating between two consecutive frames. It subtracts the second frame from the first image frame by using image subtraction operator to get the output. The problem with frame differencing is that it cannot get the complete contour of the object.

B. Temporal Differencing

This is also a difference method but the difference from framing differencing is it uses pixel wise difference method among two successive frames to detect the moving object. The drawback is this method is useless when the object is moving too fast or too slow. Because moving too slow causes minor difference between two frames causing the object to be lost and trailing regions are wrongly identified due to fast movement which is known as ghost region.

C. Optical Flow

Optical flow presents an apparent change of a moving object’s location or deformation between frames. Optical flow estimation yields a two-dimensional vector field, i. e. , motion field, that represents velocities and directions of each point of an image sequence. But this approach faces different kinds of problems due to brightness and velocity.