25-10-2016, 09:52 AM
1460963630-projectpaper.docx (Size: 16.42 KB / Downloads: 6)
Abstract- In real world applications, securing video is more important due to the happening of unusual events. Moving object detection and tracking is difficult in low resolution video and it became a challenging task due to loss of discriminative detail in visual appearance. The existing methods uses super resolution techniques to enhance the low resolution video. But these methods are not economical. The cost further increases for event detection. In this paper we present an algorithm to detect unusual events without using any super resolution techniques and it is useful for security purpose where low resolution cameras are used due to low cost. Proposed algorithm uses rolling average background subtraction technique to detect foreground object from dynamic background. This approach uses close morphological operation with structuring element in pre-processing step. Proposed algorithm is able to detect unusual events such as overcrowding or fight in low resolution video by using statistical property, standard deviation of moving objects. It process low resolution frames, so this is fast and helpful in video surveillance system where low resolution cameras are used. It does not use any classifier and avoids training requirement.
Keywords- Object detection, video surveillance, unusual event detection, low resolution video, background subtraction, morphological operation.
I. INTRODUCTION
In the past few decades, significant efforts are put in the field of moving object detection to make few applications reliable, robust and efficient: video surveillance, authentication system, robotics etc. as in [1]. There are many challenges which produce difficulties in the improvement of these applications. The challenges include illumination change, dynamic background, occlusion, shadow etc. as in [2]. These obstacles become more burden when we are performing detection and tracking in low resolution video. In low resolution video it is very difficult to exactly finding out the object of interest because most of the details such as visual features and primitives have been lost. It results in inefficient event detection. But there are some benefits of using low resolution video: low processing and transmission time, low storage as in [3].
Most of the conventional approaches are based on the high resolution (HR) video to extract contour as mentioned in [4] and shape as mentioned in [5] features of target. But these work on high resolution frames and cost will be more. Some approaches will use low resolution input in initial stages and later these videos will be enhanced to high resolution using super resolution techniques.
In the literature of unusual event detection, most of the methods use classifiers to identify the events and does not use low resolution input. These classifiers require learning time and careful attention to train the dataset. Some approaches require manual setup initially in the automated event detection system and have high computational cost. From literature we came to know we need an algorithm which is capable to detect uncommon events in low resolution video without human intervention.
This paper presents an algorithm which is able to detect abnormal events in low resolution video. This approach will enhance the security by using low resolution camera. It uses rolling average background subtraction technique to segment the foreground object with dynamic background and preserves object features simply by the application of morphological operation with structuring element. There is no need of using any classifier and training dataset. It uses only statistical property standard deviation of centroids of the blobs to identify the unusual events.
II. VIDEO SURVEILLANCE SYSTEM
Object tracking system includes four main building blocks to automate the surveillance system:
• Moving object detection
• Object tracking
• Event recognition
• Object identification
A. Moving object detection:
Detecting changes in image sequence is gaining popularity due to large number of applications in several areas. Video surveillance is one of the most important applications among them which will identify the changes in scene. There are different methods that are used to detect changes. These approaches are classified as follows:
• Frame differencing.
• Background modelling and subtraction.
1. Frame differencing
In this method difference of pixels is calculated for two or more consecutive frames in a video to detect the moving regions. Difference is calculated by setting the adaptive threshold to get region of changes. This method does not work when the object has uniform texture or move slowly. If the object stops moving this method fails. Let Pn (t) be the graylevel intensity value at the ‘t’ pixel position. Let ‘D’ be the threshold value.
|Pn (t) – Pn-1 (t)| > D
If the frame difference value is greater than the threshold value then it is a foreground pixel. Similarly if it is less than the threshold it is a background pixel. This method is adaptive to static environment only. It requires less computational cost but performance is poor.
2. Background modelling and subtraction