智能广告播放与效果评估系统

发布时间：2018-03-09 07:07

本文选题：嵌入式系统　切入点：AdaBoost　出处：《青岛理工大学》2010年硕士论文　论文类型：学位论文

【摘要】：随着商业的发展,商家对投放广告所产生的效果越来越重视,广告效果评估与智能播放是两个重要的研究内容。本文设计了一套智能广告播放与效果评估系统,通过采集广告受众的人脸图像,对人脸进行检测、跟踪与识别,判断其所属的消费群体,进而播放相应的广告并统计效果评估参数。本系统的目标是设计一个嵌入式处理终端,自动实现以上功能。硬件平台采用ADI公司生产的Blackfin系列DSP,算法主要包括人脸检测、跟踪和人脸分类,然后根据处理结果进行统计分析和智能播放,相关算法在PC机上进行了开发和调试,最后给出了向DSP进行移植的方法。在前人研究的基础上,本文主要针对人脸检测、人脸跟踪和人脸识别三个问题进行了研究: 1、在AdaBoost人脸检测算法的基础上,融合了肤色检验算法以降低误检率。AdaBoost算法主要利用人脸灰度图像信息,在检测过程中易将非人脸区域误识别为人脸区域,采用肤色匹配策略,将不在人体肤色范围内的检测结果去除,降低了误检率。 2、在粒子滤波理论框架下,讨论了基于粒子滤波的目标跟踪算法。从目标运动模型、目标观测模型、粒子重采样等几个方面分别介绍了跟踪算法的具体实现方法。并且,针对粒子滤波算法的复杂性,提出了结合Mean Shift算法的粒子采样策略,取得了良好的跟踪效果。 3、采用支持向量机对人脸进行分类。基于SVM原理,利用LIBSVM库,通过前期参数设计及正反样本训练,将人脸分为男女两类,取得了较好的分类效果。
[Abstract]:With the development of business, merchants pay more and more attention to the effect of advertising. The evaluation of advertising effect and intelligent play are two important research contents. This paper designs a set of intelligent advertising play and effect evaluation system. By collecting the face images of the advertising audience, this paper detects, tracks and recognizes the faces, judges the consumer groups to which they belong, and then plays the corresponding advertisements and calculates the evaluation parameters of the effects. The goal of this system is to design an embedded processing terminal, which can realize the above functions automatically. The hardware platform adopts Blackfin series DSP produced by ADI Company. The algorithm mainly includes face detection, tracking and face classification. Then the statistical analysis and intelligent playback are carried out according to the processing results. The related algorithms are developed and debugged on the PC. Finally, the method of porting to DSP is given. On the basis of previous studies, this paper mainly focuses on three problems: face detection, face tracking and face recognition. 1. Based on the AdaBoost face detection algorithm, the skin color detection algorithm is fused to reduce the false detection rate. AdaBoost algorithm mainly uses the face gray image information. In the process of detection, the non-face region is easily recognized as the face region, and the skin color matching strategy is adopted. The detection results which are not within the range of human skin color are removed and the false detection rate is reduced. 2. Under the framework of particle filter theory, this paper discusses the target tracking algorithm based on particle filter, and introduces the specific implementation methods of the tracking algorithm from several aspects, such as target motion model, target observation model, particle resampling and so on. Aiming at the complexity of particle filter algorithm, a particle sampling strategy combined with Mean Shift algorithm is proposed, and a good tracking effect is obtained. 3. Based on the principle of SVM, the face is classified by support vector machine (SVM). The face is classified into male and female by using LIBSVM library and the design of pre-parameters and the training of positive and negative samples, and the classification effect is good.
【学位授予单位】：青岛理工大学
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：TP391.41

【参考文献】