Aankh

SETUP :

STEP 1 : Create a virtual environment in model folder

Change working directory to model :

cd model

Install virtualenv package if it is not installed already :

pip install virtualenv

Create virtual environment :

virtualenv venv

STEP 2 : Activate virtual environment :

venv\Scripts\activate

This command need administrative permissions. Use command prompt to execute this Will not work in powershell

STEP 3 : Install required libraries

pip install -r requirements.txt

STEP 4 : Run flask app

python app.py

After this the api will be up and running on : localhost:8080 You can now give api calls from react frontend.

API ROUTES :

‘/predict_pose’ : To predict head pose
‘/predict_people’ : To predict number of people in input image

NOTE :

Both of these function takes as input a b64 encoded string which is sent via react webcam.
For testing purposes you will find a sample encoded string of foward pose image given below in sample_input.txt file .

Results of face pose estimation model :

LEFT POSE :

left_pose

RIGHT POSE :

right_pose

FORWARD :

straight_pose

NOTE :

Here we are detecting only 3 directions (forward, left , right) because that is all we need.

The number of detected directions can be increased.

To detect more directions go to draw annotations function in pose_estimator.py file and elif conditions for more angles.

if angles[1] < -15:
    GAZE = "Looking: Left"
elif angles[1] > 15:
    GAZE = "Looking: Right"
else:
    GAZE = "Forward"

Results of crowd couting model :

SAMPLE1 :

first_image

SAMPLE2 :

seccond_image

NOTE :

The number of people detected can be changed via lowering the threshold To do that go to predict funtion in app.py and change the variable threshold Our default value is 0.5

threshold = 0.5

Original repo for pose-estimation-model 👉 REPO_LINK
Link for object detection model 👉 MODEL_LINK

This site is open source. Improve this page.