Moving object detection based on unified model
- PDF / 2,025,689 Bytes
- 16 Pages / 595.276 x 790.866 pts Page_size
- 47 Downloads / 262 Views
ORIGINAL RESEARCH
Moving object detection based on unified model Anjanadevi Bondalapati1 · S. Nagakishore Bhavanam2 · E. Srinivasa Reddy1 Received: 29 February 2020 / Accepted: 30 May 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Moving object detection is an essential step in several computer visions like salient object detection, visual object tracking, and video surveillance etc. Many existing methods have a drawback of low efficiency in the challenging scenes like dynamic background, camera jitter, and bad weather images. In this research, Unified model (Yolov3-Improved Non-Maximum Suppression (INMS)) method is proposed to increases the performance in moving object detection. The Change Detection net (CDNET) dataset was trained and also COCO 2014 and PascalVOC data sets were applied to analyse the performance of the developed model. The experimental analysis shows that the developed method has higher efficiency in detecting objects in camera jitter and dynamic background scene. The performance evaluation has been done using the precision, recall, IoU and the accuracy metrics for the proposed model. The results show that the developed model has the ability to effectively identify multiple objects in the dynamic background, while the existing method has the capacity to identify only single object. Keywords Camera jitter · Dynamic background · Moving object detection · Non-maximum suppression and unified model (Yolov3—improved non-maximum suppression (INMS))
1 Introduction The change detection algorithm is important in the high level surveillance application to find the region of interest in video. Most of the “foreground” objects move intermittently (e.g. cars), as that may not be focal in the camera and uninterested background object (e.g. swaying tree branches) also tends to move (St-Charles et al. 2014). Moving object detection method should have the capacity to effectively perform in tremendous amount of available video data. Most video files consist of redundant information like dynamic background that requires more amount of storage and
* Anjanadevi Bondalapati [email protected]; [email protected] S. Nagakishore Bhavanam [email protected] E. Srinivasa Reddy [email protected] 1
Department of Computer Science and Engineering, Acharya Nagarjuna University, Nagarjunanagar, Guntur, Andhra Pradesh, India
Department of Electronics and Communication Engineering, Acharya Nagarjuna University, Nagarjunanagar, Guntur, Andhra Pradesh, India
2
computational resources (Babaee et al. 2017). The important process in surveillance method is to identify moving objects and distinguish the foreground and background object in the video. Many researches have been conducted to detect the moving object detection in a scene with more accuracy (Chen et al. 2018). Most of the methods fails to process in the challenging scenario such as obstructions, camouflage, illumination changes and shadows of the foreground object in this field. Apart from that, the moving camera lead
Data Loading...
