MTCNN-KCF-deepSort：driver face detection and tracking algorithm based on cascaded kernel correlation filtering and deep-SORT Hunan University

Format:

Book

Conference/Event

Author/Creator:

liao, Jiacai, author.

Contributor:

Cao, Libo

Wang, Qiuli

Xia, Jiahao

Zhang, Yiting

Conference Name:

WCX SAE World Congress Experience (2020-04-21 : Detroit, Michigan, United States)

Language:

English

Physical Description:

1 online resource cm

Place of Publication:

Warrendale, PA SAE International 2020

Summary:

The driver's face detection and tracking method in the fatigue driving detection system is important to Advanced Driver Assistance Systems (ADAS) and autonomous driving in various situations. The deep-SORT algorithm has integrated appearance information, motion model and the intersection-over-union (IOU) distance methods, and applied to face tracking, but it completely depends on detection information at every frame, once the detection information lacks, it will wait until the target detect bounding boxes appear again, even if the target is not disappeared or shielding. Hence, we propose to use a new tracker that doesn't depend on the detection algorithm to cascade with the deep-SORT algorithm to realize stable driver's face tracking base on detection. At First, driver's face detection and tracking will be accomplished by MTCNN-deep-SORT algorithm, Multi-task convolutional neural network (MTCNN) will complete the driver's face detection, and the detect face bounding boxes will be transferred into deep-SORT tracking algorithm, at this step, we will get the driver's face detection and tracking bounding boxes. Then, the detection bounding boxes are transferred to the kernel correlation filtering (KCF) tracking algorithm. When the driver's face is not detected, the KCF tracker will use the last frame detection information to track the driver's face at the current frame and transfer the tracking information into deep-SORT, and the deep-SORT will re-identification the driver's face in the current frame. In the case that the detection is not lost, we only record the detection information in the KCF tracker and do not start the tracker. So, we have successfully cascaded the deep learning tracking method and the traditional feature tracking method. Volunteer experiments on 20th drivers show that our method achieve more stable driver face detection and tracking when the driver's head posture changes greatly, and the cascaded tracking algorithm will improve the driver's face detection accuracy

Notes:

Vendor supplied data

Publisher Number:

2020-01-1038

Access Restriction:

Restricted for use by site license