A ViBe Based Moving Targets Edge Detection Algorithm and Its Parallel Implementation
- PDF / 1,916,104 Bytes
- 19 Pages / 439.37 x 666.142 pts Page_size
- 33 Downloads / 180 Views
A ViBe Based Moving Targets Edge Detection Algorithm and Its Parallel Implementation Han Zhang1 · Yurong Qian1 · Yuefei Wang1 · Renhe Chen1 · Chenwei Tian1 Received: 3 October 2018 / Accepted: 16 January 2019 © Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract In order to improve the computational speed and detection accuracy of the ViBe algorithm in foreground edge detection, an improved algorithm based on ViBe moving objects edge detection is proposed in this study by using the partial neighborhood model H(x, y) of a pixel at the point (x, y) to find the absolute difference between H(x, y) and the pixel’s original background model M(x, y) to determine if the pixel is moving. According to the nature of the algorithm, a method based on offset thread block coordinates to optimize thread divergence is proposed from the perspective of calculation level of kernel function and a CUDA (computing unified device architecture) based method is propsed to optimize the stream transmission between CPU and GPU in order to the run time efficiency of the new algorithm. The experimental results indicated the improved algorithm implements ghost elimination, avoids large-area irrelevant background edges and achieves better efficiency and accuracy. Keywords Parallel computing · Graphics processor · Unified computing device architecture · Moving target edge detection · ViBe algorithm · Ghost elimination · Frame difference method
1 Introduction Moving target detection has become the focus of the research in intelligent security. How to quickly and accurately extract the moving targets from the monitoring screen is the key to intelligent monitoring which lays the foundation for all the subsequent processing [1–7]. The traditional moving target detection algorithms Funded by the National Natural Science Foundation of China (61562086, 61462079), Xinjiang Uygur Autonomous Region Innovation Team XJEDU2017T002. * Han Zhang [email protected] 1
School of Software, Xinjiang University, Ürümqi 830008, China
13
Vol.:(0123456789)
International Journal of Parallel Programming
mainly include optical flow method, frame difference method and background difference method [8]. The optical flow method is basically to understand the pattern of apparent motion of image objects between two consecutive frames caused by the movement of the objects or camera. The optical flow forms 2D vector field where each vector is a displacement vector showing the movement of points (pixels) from one frame to its subsequent frame. The computational complexity of this method is high, and therefore it is not suitable for real-time processing [9]. The frame difference method obtains an image of a moving target by performing a difference operation on adjacent images, but it has the disadvantage when the moving target moves at a slow speed, it is easy to have a cavity inside the detected object and may lose the edge contour of the moving target [10]. The background difference method is to obtain the moving target by using the difference b
Data Loading...
