🕴️

Person ReID with the Python SDK

Created

Jul 3, 2020 5:36 AM

Overview

A people tracker which can keep a track of detected people and can re-identify them after occlusions or if they move outside the video frames and reappear later. This guide shows how to use it and includes a demo video

Steps

Installing dependencies and the python wheel

# copy wheel url from 
# https://github.com/getchui/offline_sdk/releases
# and run
# pip install [wheel_url]

# for example for Linux:
pip install https://github.com/getchui/offline_sdk/releases/download/0.6.6/trueface-0.6.6-cp37-cp37m-linux_x86_64.whl

2. Download Models

Download and unzip the tracker models which include a demo video file test_video.mp4

reidentification/
|--reidentify.py
|--reid.weights
|--track_detector.weights
|--test_video.mp4

Create a file called reidentify.py inside the re-identify folder the zip file just created.

3. Import SDK and video stream classes

from trueface.reid_tracker import VideoTracker
import cv2

4. Initialize VideoTracker instance

vd = VideoTracker(detector_weight_path='./track_detector.weights',
                  reid_weight_path='./reid.weights',
                  license='<your sdk token from creds.json>')

5. Start Video Capture session

video_source = './test_video.mp4' #The path where your video file resides
#Use video_source = 0 if you want to run tracker on your webcam
cap = VideoCapture(video_source)

6. Create main loop

frame_id = 0
while cap.isOpened():

    frame_id += 1
    ret, img = cap.read()

    if ret:
        # We do not need to process every frame for tracking
        if not frame_id % 3:
            ids, box, class_conf = vd.run(img)
            # Print ids so you can visually verify detected ids
            print(ids)
            # If person is detected, draw bounding box around the detection
            if len(ids) != 0:
                img = vd.draw_box(image=img, bbox=box, identities=ids)
            # Show output videostream
            cv2.imshow('Trueface.ai', img)
            if cv2.waitKey(33) == ord('q'):
                break
    else:
        break

Full Code

from trueface.reid_tracker import VideoTracker
import cv2

vd = VideoTracker(detector_weight_path='./track_detector.weights',
                  reid_weight_path='./reid.weights',
                  license='<your sdk token from creds.json>')

video_source = './test_video.mp4' #The path where your video file resides
#Use video_source = 0 if you want to run tracker on your webcam
cap = VideoCapture(video_source)

frame_id = 0
while cap.isOpened():

    frame_id += 1
    ret, img = cap.read()

    if ret:
        # We do not need to process every frame for tracking
        if not frame_id % 3:
            ids, box, class_conf = vd.run(img)
            # Print ids so you can visually verify detected ids
            print(ids)
            # If person is detected, draw bounding box around the detection
            if len(ids) != 0:
                img = vd.draw_box(image=img, bbox=box, identities=ids)
            # Show output videostream
            cv2.imshow('Trueface.ai', img)
            if cv2.waitKey(33) == ord('q'):
                break
    else:
        break
cap.release()

Video Demo

🕴️Person ReID