Struck跟踪算法:Structed Output Tracking with Kernels (ICCV /2011年)
原理: Adaptive tracking-by-detection methods
最近做项目,需要借鉴一下这个算法,于是就打算好好学习这个算法。
算法下载地址:http://download.csdn.net/detail/sunboyiris/7681943
首先介绍一下如何调通这个算法,首先要引入两个库:Eigen库和OpenCV库
Eigen库配置:
Eigen库下载地址:http://eigen.tuxfamily.org/index.php?title=Main_Page#Download
解压后对VS操作如下:
然后调用其库函数就OK了。
OpenCV库配置:
见前面文章
、
要看struct跟踪算法的效果,首先要看的是config.txt文件
# quiet mode disables all visual output (for experiments).
quietMode = 0
# debug mode enables additional drawing and visualization.
debugMode = 1
# base path for video sequences.
sequenceBasePath = sequences //路径
# path for output results file.
# comment this out to disable output.
#resultsPath = log.txt
# video sequence to run the tracker on.
# comment this out to use webcam.
#sequenceName = girl
sequenceName = girl //调用算法提供的图像序列
# frame size for use during tracking.
# the input image will be scaled to this size.
frameWidth = 320
frameHeight = 240
# seed for random number generator.
seed = 0
# tracker search radius in pixels.
searchRadius = 30
# SVM regularization parameter.
svmC = 100.0
# SVM budget size (0 = no budget).
svmBudgetSize = 100
# image features to use.
# format is: feature kernel [kernel-params]
# where:
# feature = haar/raw/histogram
# kernel = gaussian/linear/intersection/chi2
# for kernel=gaussian, kernel-params is sigma
# multiple features can be specified and will be combined
feature = haar gaussian 0.2 //haar gaussian系数设置
#feature = raw gaussian 0.1
#feature = histogram intersection
在此就配置好了参数,看一下运行效果: