Ask Your Question
0

Identify people (not from face?)

asked 2018-07-02 17:11:21 -0500

Hello World,

This is my first post, I am very new to OpenCV, so any input will be very valuable to me. I was wondering if there was a way to detect a person from behind? Let's say I was skydiving and I wanted to see how many people were ahead of me (and if I can somehow find out WHO was ahead of me, even better), just from the back with their suits on...

I was thinking of placing various color Xs on the back of their suits, each color meaning a different person, and I can make the Xs out of duct tape or magnetic tape.

I am not sure if this is making sense. I'm trying hard to explain but I am very tired, feel free to ask qualifying clarifying questions or anything like that. Thanks for taking time out of your day to read this! :)

Cheers, Michelle

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted
0

answered 2018-07-02 23:17:24 -0500

dkurt gravatar image

@sparklegirl2929 You may be interested in person reidentification algorithms which can answer if two people from different images are the same person or not depends on their appearance but not only on face. As an example, there are models trained by Intel: https://software.intel.com/openvino-t... (search for "Identify Someone in Different Videos" topic).

See https://github.com/opencv/opencv/wiki... if you want to use OpenCV to import these models.

edit flag offensive delete link more

Comments

Thanks! ) I will look into it. What about identifying the number of people in the video/picture, not so much who it is... that would just be icing on the cake.

Thanks!

sparklegirl2929 gravatar imagesparklegirl2929 ( 2018-07-03 09:23:15 -0500 )edit

Traditional image identification systems have two phases

  1. Object Detection - Tell you where the object is (x,y,w,h) coordinates
  2. Object Identification - Tell you how similar two object are - its feed with images from the bounding boxes from before

So if you just need to detect faces / persons. Just uses a pre trained model (yolo, ssd) for this.

holger gravatar imageholger ( 2018-07-04 05:18:02 -0500 )edit

That is a good idea, thank you! What is yolo and ssd?

Thanks again :)

sparklegirl2929 gravatar imagesparklegirl2929 ( 2018-07-09 11:50:29 -0500 )edit

I would call these network / model architectures.

  • yolo - You only look once - a method for extracting bounding boxes in one forward (fast)
  • ssd- single shot device - similar to yolo
holger gravatar imageholger ( 2018-07-10 01:32:33 -0500 )edit
Login/Signup to Answer

Question Tools

1 follower

Stats

Asked: 2018-07-02 17:11:21 -0500

Seen: 465 times

Last updated: Jul 02 '18