NCTU

Tag and Track for Long-term person tracking and re-identification

2017/08/02
CodeX17-YKW-00-1
Professorprofessor Yuan-Kai Wang
ApplicationSurveillance and security
FunctionImage recognition and Object/human tracking
Technical BenefitExpand monitoring scope, Improve automatic level, Improve system integration, Improve system intelligence, and Working on multi-camera at the same time
Technology StatusCan be transferred

In recent years, the government and the public attach great importance to social security, but due to the large number of the surveillance cameras, management and monitoring by human is becoming less feasible, so the intelligent surveillance system emerges.

Intelligent surveillance system is the future direction of social security. The purpose of intelligent surveillance system is to manage large surveillance network, and there will be some challenges such as object detection in different surroundings, multi-object tracking in complex environment, and object comparison in non-overlapping cameras. In order to save human resources, key frame extraction is also included in the challenges.

The proposed system, Tag and Track System(TnTs), is an intelligent surveillance system. TnTs is consist of four subsystems, person detection, multi-object tracking, person re-identification, and key frame extraction. By those subsystems, the above challenges can be solved and can achieve non-overlapping cameras tracking on specific person, also provide key information automatically.

Fig. (a)~(d) Non-overlapping cameras tracking. (a) Input image. (b)~(d) Non-overlapping cameras tracking result. Green bounding box is object detection result, red number is [camera id]_[person id].

Fig. (e)~(g) Key frame extraction. Blue bounding box is object detection result, green bounding box is face detection result, and green number is the score of key frame.

Spec. request

platform: Ubuntu 14.04 LTS 64-bit

tools:C/C++、Python 2.7、Matlab R2014b or later

hardware specification:

Processor: Intel Xeon® CPU E5-1620 v2 @ 3.70GHz * 8

Memory: 60GB

Graphics: Nvidia Tesla K20

system demand: Cuda 7.5、cuDNN V3 and V4、OpenCV 3.2、Caffe