基于安卓平台的行人检测

发布时间：2018-03-23 03:34

本文选题：行人检测　切入点：LBP-HOG联合特征　出处：《内蒙古大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着社会的进步,计算机行业也持续繁荣发展。计算机智能视觉技术俨然成为一个热门的研究方向,其中行人检测技术广受研究者们青睐。实际生活中,行人检测技术具有非常广泛的用途,如致力于改善交通安全问题的车辆辅助系统,当今社会无处不在的视频监控系统,备受重视的人群数目预测与管理等,在机器人与高级人机交互等领域也有着很好的应用价值。随着Android系统在移动端的广泛应用,它的发展前景广受各界人士看好。另外,多种移动端电子产品的体积小、便于携带、性能良好等优点吸引了大量的用户,所以,Android系统在移动端的应用市场十分广阔,因此将行人检测技术移植到Android平台上有很大的必要性并且意义深远。将基于OpenCV的行人检测移植到Android平台,这不仅使用户能够方便快捷的处理视频图像,还拓宽了 Android平台的应用范围。本文主要考虑到行人检测与安卓平台具有良好的发展前景和广阔的应用领域,所以在安卓平台上实现行人检测。论文的主要工作内容罗列如下:1.对Android系统进行概述,包括Android系统的诞生发展与繁荣。详细描述了 Android系统的组成框架与结构、生命周期、四大组件和应用进程的工作过程等。2.详细介绍了行人检测技术中的基本行人特征,如SIFT特征、HOG特征与LBP特征。学习了传统的训练行人检测分类器的算法,如SVM算法。另外,研究了利用Adaboost框架训练级联分类器的整个过程,并将此运用于本次实验中。此后将因扫描多尺度图片产生的多个矩形框利用并查集技术进行融合。3.选择INRIA数据集作为本项目实验的训练集与测试集,另外测试集还包括现场拍照与加载本地图片。考虑到HOG单一特征的不足和LBP特征对它的性能的一些补充作用,本项目结合LBP和HOG两个特征来表征行人。由于联合特征的维数增加,利用SVM分类器进行行人检测时间消耗较大,所以使用Adaboost算法训练级联分类器,降低了算法的复杂性增加了行人检测的正确率。4.搭建实验编码的环境,首先搭建Android系统环境,在Android平台上引入OpenCV库,将PC端的基于OpenCV开发的行人检测移植到Android平台上。
[Abstract]:In recent years, with the progress of society, the computer industry has continued to flourish and develop. Computer intelligent vision technology has become a hot research direction, among which pedestrian detection technology is widely favored by researchers. In real life, Pedestrian detection technology has a wide range of applications, such as the vehicle assistant system which is dedicated to improving traffic safety, the ubiquitous video surveillance system in today's society, and the number prediction and management of the population that has received much attention, and so on. With the wide application of Android system in mobile terminal, its development prospect is widely appreciated by people from all walks of life. In addition, many kinds of mobile end electronic products are small in size and easy to carry. Good performance and other advantages have attracted a large number of users, so Android system in the mobile application market is very broad, Therefore, it is necessary and meaningful to transplant pedestrian detection technology to Android platform. Transplanting pedestrian detection based on OpenCV to Android platform not only enables users to process video images conveniently and quickly. It also broadens the application scope of Android platform. This paper mainly considers that pedestrian detection and Android platform have good development prospects and wide application fields. The main work of this paper is listed as follows: 1. This paper summarizes the Android system, including the birth, development and prosperity of the Android system. It describes in detail the framework and structure of the Android system, its life cycle. The basic pedestrian features in pedestrian detection technology, such as SIFT feature hog feature and LBP feature, are introduced in detail. The traditional algorithms of training pedestrian detection classifier, such as SVM algorithm, are studied. The whole process of training cascade classifier with Adaboost framework is studied. After that, the multiple rectangular frames generated by scanning multi-scale images are fused by the technique of parallel search. 3. The INRIA data set is selected as the training set and test set of the project experiment. In addition, the test set also includes taking pictures in situ and loading local images. Considering the shortcomings of the single feature of HOG and the supplement of LBP features to its performance, This project combines LBP and HOG features to represent pedestrians. Because the dimension of joint features increases and the time of pedestrian detection using SVM classifier is large, Adaboost algorithm is used to train cascade classifier. The complexity of the algorithm is reduced and the accuracy rate of pedestrian detection is increased. Finally, the environment of experimental coding is built. Firstly, the Android system environment is built, and the OpenCV library is introduced into the Android platform, and the pedestrian detection based on OpenCV is transplanted to the Android platform.
【学位授予单位】：内蒙古大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】