Pedestrian flow funnel generation method and apparatus, storage medium and electronic device

Patent metadata

Field	Value
Publication number	US-11210795-B2
Application number	US-201816758958-A
Country	US
Kind code	B2
Filing date	Oct 24, 2018
Priority date	Oct 24, 2017
Publication date	Dec 28, 2021
Grant date	Dec 28, 2021

Abstract

The disclosure relates to the technical field of data processing. Disclosed are a pedestrian flow funnel generation method and apparatus, a storage medium and an electronic device. The method can comprise: acquiring a current frame of image, and tracking and updating, according to a multi-target tracking algorithm, a head and shoulder area in a tracking sequence set in the current frame of image; acquiring, according to a head and shoulder recognition model, a head and shoulder area in the current frame of image, and updating, according to the head and shoulder area in the current frame of image, the tracking sequence set; analysing a motion track of each head and shoulder area in the tracking sequence set so as to count pedestrians, and when the current frame of image is the last frame of image, generating a pedestrian flow funnel based on a counting result of pedestrians.

First claim

What is claimed is:

1. A method for generating a pedestrian flow funnel, comprising:
obtaining a current frame image, and tracking and updating head-shoulder regions in a tracking sequence set in the current frame image according to a multi-target tracking algorithm;
obtaining head-shoulder regions in the current frame image according to a head-shoulder recognition model, and updating the tracking sequence set according to the head-shoulder regions in the current frame image; and
analyzing a motion trajectory of each head-shoulder region in the tracking sequence set to count pedestrians, and when the current frame image is a last frame image, generating the pedestrian flow funnel based on a counting result of the pedestrians,
wherein tracking and updating head-shoulder regions in a tracking sequence set in the current frame image according to a multi-target tracking algorithm comprises:
tracking and updating the head-shoulder regions in the tracking sequence set in the current frame image according to a kernel correlation filter tracking algorithm, respectively,
wherein tracking and updating the head-shoulder regions in the tracking sequence set in the current frame image according to a kernel correlation filter tracking algorithm respectively comprises:
obtaining candidate tracking targets corresponding to each head-shoulder region in the tracking sequence set in the current frame image based on a position of each head-shoulder region in the tracking sequence set in a previous frame image, respectively;
correspondingly calculating response values of the candidate tracking targets corresponding to each head-shoulder region according to a tracker corresponding to each head-shoulder region, respectively;
determining a candidate tracking target with a largest response value among the candidate tracking targets corresponding to each head-shoulder region as a tracking target of the corresponding head-shoulder region in the current frame image; and
correspondingly updating each head-shoulder region in the tracking sequence set according to the tracking target of each head-shoulder region in the current frame image.

2. The method for generating a pedestrian flow funnel according to claim 1 , before obtaining a current frame image, the method further comprises:
obtaining a first frame image, obtaining head-shoulder regions in the first frame image according to the head-shoulder recognition model, and initializing the tracking sequence set through the head-shoulder regions.

3. The method for generating a pedestrian flow funnel according to claim 2 , wherein generating the pedestrian flow funnel based on a counting result of the pedestrians comprises:
generating the pedestrian flow funnel based on the counting result of the pedestrians and in combination with the age groups and gender of each head-shoulder region in the tracking sequence set.

4. The method for generating a pedestrian flow funnel according to claim 1 , wherein the method further comprises:
identifying age groups and gender of each head-shoulder region in the tracking sequence set according to a gender-age recognition model.

5. The method for generating a pedestrian flow funnel according to claim 4 , wherein the method further comprises: constructing the gender-age recognition model, comprising:
obtaining a third training sample set marking gender and age in a LFW dataset and a social network site; and
training a gender-age network by using the third training sample set to obtain the gender-age recognition model, wherein the gender-age network comprises three convolutional layers and three fully connected layers.

6. The method for generating a pedestrian flow funnel according to claim 1 , wherein the method further comprises: calculating the tracker corresponding to each head-shoulder region, comprising:
obtaining a first training sample set corresponding to each head-shoulder region in the tracking sequence set in the previous frame image based on the position of each head-shoulder region in the tracking sequence set in the previous frame image, respectively; and
training a regression model according to the first training sample set corresponding to each head-shoulder region respectively, to obtain the tracker corresponding to each head-shoulder region.

7. The method for generating a pedestrian flow funnel according to claim 1 , wherein the method further comprises: generating the head-shoulder recognition model according to a convolutional neural network, comprising:
training a MobileNet network according to an ImageNet categorical dataset to obtain a weight value of the MobileNet network;
adding convolutional layers with a preset number of layers above the MobileNet network to obtain a head-shoulder detection network, wherein a size of the convolutional layers decreases layer by layer; and
obtaining a second training sample set marked with the head-shoulder regions, and training the head-shoulder detection network retaining the weight value of the MobileNet network according to the second training sample set, to obtain the head-shoulder recognition model.

8. The method for generating a pedestrian flow funnel according to claim 1 , wherein updating the tracking sequence set according to the head-shoulder regions in the current frame image comprises:
calculating similarity between each head-shoulder region in the current frame image and each head-shoulder region in the tracking sequence set; and
updating the tracking sequence set according to the similarity.

9. The method for generating a pedestrian flow funnel according to claim 8 , wherein calculating similarity between each head-shoulder region in the current frame image and each head-shoulder region in the tracking sequence set comprises:
calculating similarity between each head-shoulder region in the current frame image and each head-shoulder region in the tracking sequence set according to a following equation:

sim
⁡

(

Q
i

,

Q
j

)

=

Q
i

⁢
area

⋂

Q
j

⁢
area

Q
i

⁢
area

⋃

Q
j

⁢
area

wherein, sim(Q i , Q j ) is similarity between the i-th head-shoulder region Q i the current frame image and the j-th head-shoulder region Q j in the tracking target set, Q i area is an area of the i-th head-shoulder region Q i in the current frame image, and Q j area is an area of the j-th head-shoulder region Q j in the tracking sequence set.

10. The method for generating a

Assignees

• Beijing Jingdong Shangke Information Technology Co Ltd
• Beijing Jingdong Century Trading Co Ltd

Inventors

• Che Guangfu
• Chen Yu
• An Shan
• Liu Qiang
• Weng Zhi

Classifications

• G06V10/62

  relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking · CPC title

• G06T7/246 Primary

  using feature-based methods, e.g. the tracking of corners or segments · CPC title

• G06T7/248 Primary

  involving reference images or patches · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title