Abstract
Background subtraction is a prevailing method for moving object detection in videos with stationary backgrounds. However, accurate and real-time moving object detection is challenging in the presence of complex dynamic scenes. This paper presents a novel technique for background subtraction based on the dynamic autoregressive moving average (ARMA) model. Specifically, we utilize the temporal and spatial correlation of images in a video sequence to model each pixel to accurately model the background image's dynamic characteristics. In addition, we apply an adaptive least mean square (LMS) scheme to update the parameters of the background model to offset the dramatically dynamic characteristic of the background. The proposed algorithm is evaluated on two publicly available benchmark datasets with complex dynamic backgrounds. The experimental results show that this technique is robust and effective for background subtraction in complex dynamic backgrounds and is a promising moving object detection scheme for real-time visual surveillance.
Highlights
The extensive development of intelligent visual surveillance systems demands that systems possess powerful and real-time image processing ability to identify, locate and track moving objects
EXPERIMENTAL RESULTS AND ANALYSIS This section introduces the datasets for moving object detection, the parameter setup and the evaluation metrics, presents the experimental results and performance analysis
To qualitatively validate the robustness of the proposed algorithm, we select the following representative videos from the above datasets that cover a large number of challenging surroundings, as shown in Table 3: (1) ‘‘waving trees’’ (WT) depicts a scenario with trees shaken by the wind; (2) ‘‘time of day’’ (TOD) shows a gradual illumination change; (3) ‘‘badminton’’ (BAD) is recorded in an outdoor environment with camera jitter; (4) ‘‘camouflage’’ (CAM) shows a scenario where the color of an object is similar to the background color; (5) ‘‘skating’’ (SKT) presents the challenge of poor winter weather conditions; (6) ‘‘pets2006’’ (PET) presents a scenario known for intermittent object motion; and (7) ‘‘library’’ (LIB) is a series of infrared thermal images
Summary
The extensive development of intelligent visual surveillance systems demands that systems possess powerful and real-time image processing ability to identify, locate and track moving objects. Inspired by pixel-based and region-based algorithms, we propose a simple but robust moving object detection scheme based on a dynamic ARMA model to overcome these difficulties. The SuBSENSE algorithm incorporates local binary similarity patterns descriptors to model the background [23] and is robust to noise and variation in the background These featurebased methods produce good segmentation results compared to those of pixel-based methods but are not sufficiently stable for complex frequent variations in the background. Recent methods based on convolutional neural networks have been proposed for moving object detection [32]–[34] These approaches perform well in complex dynamic scenes but require a large quantity of labeled data for training.
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
