easyMTracking

Methods:

multiple target tracking, only pedestrian tracking was implemented at present!
Detection by yolov2, calman filter predict motion trajectory, and build assignment problem sovled by Hungarian algorithm .

Demo video:

Need cuda8.0 and cudnn6.0 , and opencv3.3.0 on win10,

1.Download video and yolo wieghts:

yolo-voc.weights:http://pjreddie.com/media/files/yolo-voc.weights
video: http://pan.baidu.com/s/1cwHj6u or other video.
put video file and weights file into "x64/Release" 

2. cd "x64/Release" , and cmd run multi_tracker_yolo_voc_demo.cmd

3. About paramters:

	"{ t  tracker          |0  | tracker type: 0-video sort kalman, 1- seq sort kalman | }"
	"{ d  debug           |0  | debug : 0-yes, 1-false | }"
	"{ mint  minTraceSize |8  | min trace size > 2 | }"
	"{ in   inFile        |   |input video| }"
	"{ out  outFile       |   |write processed video| }"
	"{ seq  sequences     |PETS09-S2L1 |process sequences| }"
	"{ det  detector      |0  |detector type: 0-identified_seq, 1-yolov2 |}"
	"{ net  netfile       |  |dl net file |}"
	"{ weight  weightfile |  |dl net weight file |}"
	"{ objnamefile  objectnamefile |voc.names  |data sets object names |}"
	"{ filname  filtername |person  | custom made object names like: person,car,bicycle,bus | }"

Build

1.Istall win10 vs2015, cuda8.0 and cudnn6.0
2.Install opencv3.3.0 ,here i used opencv-3.3.0-vc14.zip, link:https://sourceforge.net/projects/opencvlibrary/files/opencv-win/3.3.0/opencv-3.3.0-vc14.exe/download
3.opencv open_tracker.sln, and build x64 release.

Reference materials and projects:

1.Multitarget-tracker,https://github.com/Smorodov/Multitarget-tracker
2.Simple Online and Realtime Tracking, https://arxiv.org/abs/1602.00763
3.windows yolo https://github.com/unsky/yolo-for-windows-v2