this is the loop closure testing for BRLD/SIFT by Shiwei Zhuang This program is all in matlab ###put data directory to BOW_LIBSVM

• The local data only has the BRLD descriptors, SIFT descriptors is too big to upload, you can generate SIFT descriptors  by your SIFT,of course you should read the code and  remove some ".mat" files in data\global.

###1.Open the directory “labels”

• open ini.m,change the path pre_data_path='E:\loop_closure_kyle\BOW_LIBSVM\images' to your own path
• run prepare_training.m
• then run prepare_testing .m

###2.Go back to the main directory

• open ini.m (this ini.m is different from the former one),change the path "rootpath='E:\loop_closure_kyle\BOW_LIBSVM'" to your own path;
• run main.m, then you can see the matching results

if there is an error "未定义函数或变量 'pg_opts'", change your matlab path to the main directory BOW_LIBSVM ###3.This version is for BRLD testing , the binary robust line descriptor if you want to see the results of SIFT, run the code which is annotated in main.m

##各个函数说明 ###ini初始化： 规定pg_opts的各种路径
创建文件夹，data/local和global
Local:BRLD和SIFT两个文件夹，用于存储每张图片的描述子
Global:用来存储一些全局flag ##GenerateBRLDdescriptors 生成BRLD的描述子,存储在data/local/BRLD的每个文件夹里
flag：data/global/desBRLD_settings，标志BRLD是否生成
如果要添加训练/测试图片，改变后将desBRLD_settings.mat移除，修改循环变量，就可以为新添加的图片生成描述子
error：lsd无法调用，重新编译lsd，见文件夹lsd，lsd在matlab中的编译及使用

###CalculateBRLDDictionary 通过k-means聚类的方法，用training中的图片来生成vocabulary
600个聚类中心,是一个600x128
结果Save到data/global/BRLD_dictinary.mat
训练照片所有线条描述子的聚类中心就是词典中的词汇
flag：BRLD_features_settings

##do_assignment 分配tf-idf权重，即计算每张图片的BOW向量 flag：BOW_BRLD_settings 如果添加了新图片，要生成bow向量，把flag从global文件夹里移除， ##do_index 找到测试图像和训练图像之间的相似性，并获得最相似的k个图像
计算相似性用的是subspace函数