Cauchy Mixture Model-based Foreground Object Detection with New Dynamic Learning Rate Using Spatial and Statistical info
- PDF / 4,886,179 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 38 Downloads / 131 Views
RESEARCH PAPER
Cauchy Mixture Model-based Foreground Object Detection with New Dynamic Learning Rate Using Spatial and Statistical information for Video Surveillance Applications D. Sowmiya1
•
P. Anandhakumar1
Received: 7 December 2017 / Revised: 2 May 2019 / Accepted: 20 May 2019 The National Academy of Sciences, India 2019
Abstract This paper presents a background modeling technique based on Cauchy mixture model (CMM) for moving object detection. The proposed approach detects the objects under different scenarios such as cluttered background, sudden and varied illumination, in the presence of shadow, slow- and fast-moving objects and under different weather conditions. The qualitative and quantitative performance evaluation of the proposed method on background model challenge (BMC) datasets, containing both real and synthetic videos, exhibits a superior performance over the state-of-the-art methods. The average accuracy for real and synthetic videos is 94.92% and 98.01%, respectively. The average F-score for real videos is 96.49%, and that for synthetic videos is 98.86%. The area under the curve (AUC) reveals an improved performance of 4.6% and 3.6% for real videos and synthetic videos of BMC dataset, respectively. Keywords Cauchy mixture model Object detection Video surveillance Background modeling
1 Introduction Moving object detection is a primary step in most of the computer vision applications such as human activity recognition, gesture recognition for human–machine interface and visual surveillance of vehicles and pedestrians. Visual surveillance-based monitoring system mounts static surveillance cameras in public environment to & D. Sowmiya [email protected] 1
Anna University Chennai, Chennai, Tamilnadu, India
capture the environmental occurrence. This involves various kinds of object that may be static or dynamic present in the foreground and background region [1]. One of the requirements for any computer vision application is to separate the region of interest (human, vehicles, gestures, etc.) from the surrounding environment. A background model is generated using the background pixels that do not change for a long period of time. Then, the object detection is performed by comparing the changing foreground pixels with the nominal static background pixels with an appropriate thresholding level. The simple pixel changes alone are not sufficient for efficient moving object detection. The other factors to consider are noise, illumination fluctuations (both local and global), weather (cloudy, sunny, foggy), visual effects such as shadows and reflections. A robust moving object detection method should be capable of handling issues such as dense vegetation in the background, different ground type, casted shadows, continuous movement of objects (vehicle, pedestrian) in surveillance environment, different climatic conditions, gradual and sudden illumination changes, the presence of varied size objects moving in diverse speed (slow or fast) [2]. A customary approach for detecting moving objects
Data Loading...
