app.py

import streamlit as st
import tempfile
import matplotlib.pyplot as plt
from scipy.signal import find_peaks
from pyampd.ampd import find_peaks_adaptive
import numpy as np
from vitals import predict_vitals,hear_rate,load_model
import heartpy as hp
from process import remove_outliers,fourier_analysis
from PIL import Image
img=Image.open('logo.png')


col1, mid, col2 = st.beta_columns([1,1,20])
with col1:
    st.image('logo.png', width=60)
with col2:
    st.title('Calculate heart rate from video')


@st.cache
def loadmodel():
	model=load_model()
	return model

model=load_model()

#@st.cache
uploaded_file = st.file_uploader("Upload Files",accept_multiple_files=False,type=['mp4','avi'])#upload file
if uploaded_file is not None:#check if file is present
	with st.spinner('Wait for it...'):
		tfile = tempfile.NamedTemporaryFile(delete=False) 
		tfile.write(uploaded_file.read())
		pulse,resp,fs,sample_img=predict_vitals(tfile.name,model)
	st.success('Done!')
	np.savetxt('pulse.txt',pulse)
	st.write('frame rate',np.round(fs,3))
	# d=st.slider('Select threshold', min_value=5 , max_value=15 , value=10 , step=2 )
	# st.write(d)
	# peaks1, _ = find_peaks(pulse,distance=d)
	hr_adaptive=[]
	for i in range(10,100,5):
		peaks=find_peaks_adaptive(pulse, window=i)
		hr_adaptive.append(hear_rate(peaks,fs))
	hr_adaptive=remove_outliers(hr_adaptive)
	hrAmed=np.median(hr_adaptive)
	hrAstd=np.std(hr_adaptive)
	hrAmean=np.mean(hr_adaptive)
	
	# hr=hear_rate(peaks1,fs)
	# print(hr)
	hpy=hp.process_segmentwise(pulse,sample_rate=fs,segment_overlap=0.75,segment_width=15) [1]['bpm']
	hpy=[x for x in hpy if str(x) != 'nan']
	#hpy=remove_outliers(hpy)

	hpy_single=hp.process(pulse,sample_rate=fs,) [1]['bpm']
	#st.header("heart rate is {}, {}, {}".format(hrAmean.round(),hrAmed.round(),hrAstd.round()))
	#st.header('heart rate mean is {}, and range is  {} - {}'.format(np.mean(hpy).round(),np.min(hpy).round(),np.max(hpy).round()))
	print("hear rate is",hpy_single,np.mean(hpy),hrAmean,fourier_analysis(pulse,fs)*60)
	st.header('heart rate is {}'.format(hpy_single*0.3+np.mean(hpy)*0.4+hrAmean*0.3))
	fig,ax=plt.subplots(2,1)
	ax[0].plot(pulse) 
	# ax[0].plot(peaks1, pulse[peaks1],label='threshold',marker= "x")
	# ax[0].plot(peaks, pulse[peaks],label='adaptive',marker= "*")
	ax[0].set_title('Pulse Prediction')
	# ax[0].legend()

	ax[1].plot(resp)
	ax[1].set_title('Respiration Prediction')
	fig.tight_layout()
	st.write(fig)

	st.image(sample_img)


	st.write("Note")
st.markdown(	"""
### Note
1. Camera should be right in front of face and capturing the whole face and shoulder as in image shown below
2. There should be little to no movement of camera and head
3. Lightening should be normal, not to much bright nor dark.
4. It is suggested to face light while creating video
5. Avoid talking, eating while creating video.
6. Video should be more than one minute long. 
7. FPS of camera should not be less than 15 FPS.
8. Mobile camera should not be of low mega pixels
""")
col1, col2, col3 = st.beta_columns([1,6,1])

with col1:
	st.write("")

with col2:
	st.image('frame.jpg',caption="ideal video frame",
	width=200)

with col3:
	st.write("")

st.header("Factor effecting heart rate calculation")
st.write('Effect of camera on heart rate calculation')
col1, col2, col3 = st.beta_columns([1,6,1])
with col1:
	st.write("")
with col2:
	st.image('camera_effect.png',caption="",
	width=500)
with col3:
	st.write("")

st.write('Effect of light on heart rate calculation')
col1, col2, col3 = st.beta_columns([1,6,1])
with col1:
	st.write("")
with col2:
	st.image('light_effect.png',caption="",
	width=500)
with col3:
	st.write("")

st.write('Effect of facial expression on heart rate calculation')
col1, col2, col3 = st.beta_columns([1,6,1])
with col1:
	st.write("")
with col2:
	st.image('facial_expression.png',caption="",
	width=500)
with col3:
	st.write("")

st.write('Effect of talking on heart rate calculation')
col1, col2, col3 = st.beta_columns([1,6,1])
with col1:
	st.write("")
with col2:
	st.image('talking_effect.png',caption="",
	width=500)
with col3:
	st.write("")


st.write("""Image source: Real-Time Webcam Heart-Rate and Variability
 Estimation with Clean Ground Truth for Evaluation""")