英语原文共 16 页，剩余内容已隐藏，支付完成后下载完整资料
原文
Towards improving quality of video-based vehicle counting method for traffic flow estimation
Yingjie Xia a,c, Xingmin Shi c, Guanghua Song b,n, Qiaolei Geng c, Yuncai Liu c
- College of Computer Science, Zhejiang University, Hangzhou, Zhejiang, China
- School of Aeronautics and Astronautics, Zhejiang University, Hangzhou, Zhejiang, China
- Intelligent Transportation and Information Security Lab, Hangzhou Normal University, Hangzhou, Zhejiang, China
Article info
Article history:
Received 19 August 2014
Received in revised form 15 October 2014 Accepted 29 October 2014
Available online 6 November 2014
Abstract
The traffic flow is usually estimated to evaluate the traffic state in traffic management, and vehicle counting is a key method for estimating traffic flow. With wide deployment of cameras in urban transportation systems, the surveillance video becomes an important data source to conduct vehicle counting. However, the efficiency and accuracy of vehicle counting are seriously affected by the complexity of traffic scenarios. In this paper, we employ the virtual loop method to improve the quality of video-based vehicle counting method. As details, the expectation-maximization (EM) algorithm is fused with the Gaussian mixture model (GMM) for improving the segmentation quality of moving vehicles. In addition, a restoration method is designed to remove noise and fill holes for obtaining a better object region. Finally, a morphological feature and the color-histogram are utilized to solve occlusion issues. The effectiveness and efficiency experiments show that the proposed approach can improve the vehicle segmentation result and the vehicle occlusion detection. The accuracy of vehicle counting can also be improved significantly and reach 98%.
Keywords: Gaussian mixture model; Expectation maximization; Vehicle counting; Vehicle extraction Occlusion detection
1. Introduction
Traffic flow is a key in the urban transportation system due to that its estimation is helpful in evaluating traffic state for management. And vehicle counting is a key technique for traffic flow estimation.
Traditional vehicle counting methods are mainly based on the buried inductive loop detector [1], which is an electromagnetic communication system and can detect vehicles passing at a certain point. However, such kind of equipment needs to be buried under the road surface, and therefore, is inconvenient to be maintained. Compared with the inductive loop detector, the traffic camera is installed on the roadside and extensively used in the urban intelligent transportation systems (ITS). Since the camera is easy to be operated and maintained, the traffic video data are extensively applied in vehicle counting [2], violation detection [3], vehicle tracking [4], traffic light detection [5], etc.
The video-based vehicle counting primarily depends on the vehicle detection, which has been implemented by a number of methods, such as optical flow (OF) [6], frame differencing (FD) [7] and background subtraction (BS) [8]. However, these methods always leave the problems in efficiency, accuracy, and effectiveness. Furthermore, for recognizing the object efficiently, feature correlation hypergraph (FCH), which is proposed in [9], can be used to model multiple features and enhance the recognition of vehicles. And multi-view based methods are also proposed to improve the accuracy of object classification in [10]. However, both approaches are suitable for fine-grained classification and recognition.
In this paper, a virtual loop-based method is proposed to improve the quality of vehicle counting method. In this method, BS is used to detect the moving vehicles and an improved Gaussian mixture model (GMM) with expectation-maximization (EM) algorithm is proposed for obtaining better tracking results. The main contribution of this work includes the following: firstly, an EM-GMM method is proposed to improve the quality of the resulted background image; in addition, a region restoration method is used to remove the noises and fill the holes for getting complete region contour; and finally the occlusion cases are detected and solved by using morphological features and the color histogram.
This paper is structured as follows. Section 2 reviews the methods proposed in the literatures of vehicle counting. Section 3 elaborates the proposed method, including the EM-based GMM, the restoration method and the occlusion detection method. The experimental results and analysis are given in Section 4. In Section 5, the conclusion is drawn and the future work is outlined.
2. Related work
Various kinds of approaches are proposed for improving the quality of video-based vehicle counting method. The literatures dealing with the object detection, object tracking, and vehicle counting are reviewed as follows.
2.1. Object detection and tracking
As for object detection and tracking, a number of methods are proposed, including optical flow, frame difference, and background subtraction.
Optical flow is based on the pixel-level intensity estimation and can detect moving objects without any prior information about the situati
剩余内容已隐藏，支付完成后下载完整资料
资料编号：[151532]，资料为PDF文档或Word文档，PDF文档可免费转换为Word
以上是毕业论文外文翻译，课题毕业论文、任务书、文献综述、开题报告、程序设计、图纸设计等资料可联系客服协助查找。
