摘要论文利用灰度共生矩阵算法研究杂乱场景下的人群密度估计。首先在学校选取较为简单和较为复杂两种场景，拍摄获得图像库。然后通过灰度共生矩阵提取 355 张图像的统计特征，并为其统计人数，得到 2 个训练集和 2 个验证集。分别采用最小二乘线性拟 合和局部加权线性回归两种方法对每个场景进行训练和测试：在较为简单的场景下两种 算法结果都较为理想，估算误差低；在场景较为复杂，人群密度较高时，两种算法拟合的 平均精度最低为 80%。通过对比分析，局部加权的方法在中低密度时也能保持较低的估算 误差，更加稳健和精确。最后改变局部加权方法参数值的大小，研究对人群密度估计的影 响。77753

毕业论文关键词  人群密度  灰度共生矩阵  最小二乘线性拟合  局部加权线性回归

毕 业设计 说明书 外文摘 要

Title Crowd Density Estimation Algorithm         

Abstract Crowd density estimation under a clutter environment is studied by the gray level co-occurrence matrix algorithms in this thesis。 First, Two types of videos respectively representing relatively simple and complex scenes are  collected in university campus to build a image library。 Then two sets of training and test videos can be got by extracting the statistical characteristics of 355 images through gray level co-occurrence matrix and counting the number of people in the images 。 Respectively using the least squares linear fitting method and the locally weighted linear regression method , we find that these two algorithms both do a fantastic job in relatively simple scene and the accuracy of them can lowest achieve nearly 80% in the high population density。 Through a contrary analysis, the method of locally weighted linear regression in the low density of population is found even more robust and precise and keep a relatively low estimated error rate。 Last but not least, the influence of the local weighted method to the crowd density estimation is studied by changing the values of the parameters of it 。 

Keywords  Crowd density  GLCM  LSLF  LWLR

本科毕业设计说明书 第 I  页

目 次

1  绪论 1

1。1  课题研究背景 1

1。2  国内外研究现状 1

1。3  本论文完成的工作和论文章节安排 3

2  基于灰度共生矩阵的算法原理 4

2。1  灰度共生矩阵原理 4

2。2  灰度共生矩阵的计算 6

2。3  图像掩模 8

2。4  拟合数据的准备 9

2。5  最小二乘线性拟合原理及实现 10

2。6  局部加权线性回归原理及实现 11

2。7  本章小结 12

3  实验方法及实验结果 14

3。1  实验环境 14

3。2  图像库的选取 14

3。3  实验参数的选择 15