app.py

# -*- coding: utf-8 -*-
"""

Luke Fullard: 16 June 2024

Script to track an object of interest in a video

"""

import streamlit as st
import cv2
import numpy as np
from PIL import Image
import tempfile
import pandas as pd
import os
from io import BytesIO
import base64
from skimage import filters
###############################################################################
###############################################################################
###############################################################################
# Function to apply image adjustments with new options
def apply_adjustments(frame, grayscale, contrast, blur, edges, sharpen_amount, subtract_background, lower_threshold, upper_threshold, binarize, binarize_threshold):
    if grayscale:
        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    if contrast > 1.0:
        # Clip the values to avoid wrapping around
        frame = np.clip(contrast * frame, 0, 255).astype(np.uint8)
    if blur > 0:
        frame = cv2.GaussianBlur(frame, (blur, blur), 0)
    if edges:
        frame = cv2.Canny(frame, lower_threshold, upper_threshold)

    # Sharpening
    if sharpen_amount > 0:
        frame = filters.unsharp_mask(frame, radius=1, amount=sharpen_amount)
        frame = (frame * 255).astype(np.uint8)  # Convert back to uint8

    # Background Subtraction
    if subtract_background:
        # Initialize the background subtractor
        fgbg = cv2.createBackgroundSubtractorMOG2()
        frame = fgbg.apply(frame)
        
    if binarize:
        _, frame = cv2.threshold(frame, binarize_threshold, 255, cv2.THRESH_BINARY)    

    return frame

###############################################################################
###############################################################################
###############################################################################
def part_one():
    st.subheader('Part one: upload your video file')
    with st.expander('Instructions #1'):
        st.write('Please choose a file to upload. At present only the following formats are supported: "mp4", "avi", "mov", "mkv"')
    uploaded_file = st.file_uploader("Choose a video file", type=["mp4", "avi", "mov", "mkv"])
    return uploaded_file

###############################################################################
###############################################################################
###############################################################################
def part_two(uploaded_file):
    st.subheader('Part two: adjust image and select region of interest')
    with st.expander('Instructions #2'):
        st.write('''

Part two is a two step process:     

    

a) Use the Image adjustments in the left hand side toolbar to modify the image until suitable for tracking analysis. In practice, I have found the "Binarize" tool to be the most useful for tracking. You can use the frame number selector just above the image to see how the image adjustment affects the other frames in the video sequence.   

            

b) Set the region of interest (ROI). Using the four number sliders below, choose the initial area of interest. the X,Y numbers define the top left of the rectangular area of interest, while the width and height define the rectangle dimensions. The rectangle will be drawn on the image to help guide you.   

**NOTE: The ROI is always set on the first frame, so ensure you are on that image frame when setting the ROI.**   

   

Once you are ready, click the "Start Object Tracking" button below the image.       

                 ''')
    # Save the uploaded video file to a temporary file
    tfile = tempfile.NamedTemporaryFile(delete=False) 
    tfile.write(uploaded_file.read())
    file_name, _ = os.path.splitext(uploaded_file.name)



    # Read the first frame of the video from the temporary file
    vcap = cv2.VideoCapture(tfile.name)
    total_frames = int(vcap.get(cv2.CAP_PROP_FRAME_COUNT))

    # Close and delete the temporary file
    tfile.close()

    instructions_container = st.empty()
    left_column,right_column = st.columns(2)
    
    # User input for frame number
    frame_number = st.number_input(f'Enter a frame number between 1 and {total_frames}', min_value=1, max_value=total_frames, value=1)
    if frame_number >1:
        st.warning("WARNING: When setting the region of interest (ROI) please ensure that you are defining the region based on Frame #1.")

    # Options for adjustments
    st.sidebar.header("Image adjustments")
    grayscale = st.sidebar.checkbox("Convert to Grayscale")
    contrast = st.sidebar.slider("Contrast", 1.0, 3.0, 1.0)
    blur = st.sidebar.slider("Blur", 0, 10, 0)
    edges = st.sidebar.checkbox("Edge Detection")
    # Options for edge detection adjustments
    lower_threshold = 0
    upper_threshold = 0
    if edges:
        st.sidebar.header("Edge Detection Parameters")
        lower_threshold = st.sidebar.slider("Lower Threshold", 0, 255, 100)
        upper_threshold = st.sidebar.slider("Upper Threshold", 0, 255, 200)

    sharpen_amount = st.sidebar.slider("Sharpen Amount", 0.0, 2.0, 0.0)
    subtract_background = st.sidebar.checkbox("Background Subtraction")
        
    # Options for binarization adjustments
    binarize = st.sidebar.checkbox("Binarize")
    binarize_threshold=0
    if binarize:
        st.sidebar.header("Binarization Parameters")
        binarize_threshold = st.sidebar.slider("Binarization Threshold", 0, 255, 128)
    
        # Seek to the specified frame
    vcap.set(cv2.CAP_PROP_POS_FRAMES, frame_number-1)
    success, frame = vcap.read()

    if success:
        # Manual ROI entry
        with instructions_container:
            st.write("Enter the Region of interest (ROI) coordinates")
        
        with left_column:
            x = st.number_input('Enter X coordinate of top-left corner', min_value=0, value=0)
            w = st.number_input('Enter width of the ROI', min_value=1, value=100)
        with right_column:    
            y = st.number_input('Enter Y coordinate of top-left corner', min_value=0, value=0)
            h = st.number_input('Enter height of the ROI', min_value=1, value=100)
        # Apply adjustments to the frame
        adjusted_frame = apply_adjustments(frame, grayscale, contrast, blur, edges, sharpen_amount, 
                                           subtract_background, lower_threshold, upper_threshold, binarize, 
                                           binarize_threshold,
                                           )

        # Draw the rectangle on the frame
        cv2.rectangle(adjusted_frame, (x, y), (x + w, y + h), (255, 0, 0), 2)
        # Display the adjusted frame
        st.image(Image.fromarray(adjusted_frame), caption=f'Adjusted Frame at {frame_number}', use_column_width=True)

    else:
        st.error(f"Could not read frame number {frame_number}.")
    st.write('---')
    # Release the video capture object
    # vcap.release()    
    return vcap,file_name,total_frames,grayscale, contrast, blur, edges, sharpen_amount, subtract_background, lower_threshold, upper_threshold, binarize, binarize_threshold, x,y,w,h

###############################################################################
###############################################################################
###############################################################################
def part_three(vcap,temp_dir,file_name, grayscale, contrast, blur, edges, sharpen_amount, subtract_background, lower_threshold, upper_threshold, binarize, binarize_threshold,x,y,w,h):
    st.subheader('Part Three: Your object being tracked!')
    st.write("...please be patient, it may take some time...")
    vcap.set(cv2.CAP_PROP_POS_FRAMES, 0)
    # Initialize the object tracker
    tracker = cv2.TrackerCSRT_create()
    
    #initialize positions
    x_position = []
    y_position = []
    w_position = []
    h_position = []
    
    
    
    # Read the first frame
    success, frame = vcap.read()
    if success:
        # Get video properties
        frame_width = int(vcap.get(cv2.CAP_PROP_FRAME_WIDTH))
        frame_height = int(vcap.get(cv2.CAP_PROP_FRAME_HEIGHT))
        fps = vcap.get(cv2.CAP_PROP_FPS)
        
        # Define the codec and create VideoWriter object
        fourcc = cv2.VideoWriter_fourcc(*'XVID')
        out = cv2.VideoWriter(f'{temp_dir}/{file_name}_ADJUSTED.avi', fourcc, fps, (frame_width, frame_height))
        
        # Apply adjustments to the first frame
        adjusted_frame = apply_adjustments(frame, grayscale, contrast, blur, edges, sharpen_amount, subtract_background, lower_threshold, upper_threshold, binarize, binarize_threshold)
        
        # Define an initial bounding box
        bbox = (x, y, w, h)
        
        # Initialize tracker with first frame and bounding box
        tracker.init(adjusted_frame, bbox)
        
        tracking_image_box = st.empty()
        tracking_image_info = st.empty()
        # Loop over the frames of the video
        frame_number = 0
        while True:
            # Read a new frame
            success, frame = vcap.read()
            if not success:
                break
            # Apply adjustments to the frame
            adjusted_frame = apply_adjustments(frame, grayscale, contrast, blur, edges, sharpen_amount, subtract_background, lower_threshold, upper_threshold, binarize, binarize_threshold)
            # Update tracker
            success_tracker, box = tracker.update(adjusted_frame)
            if success_tracker:
                
                # Draw the tracking box
                (x, y, w, h) = [int(v) for v in box]
                cv2.rectangle(adjusted_frame, (x, y), (x + w, y + h), (0, 255, 0), 2)
                # Write the adjusted frame to the output video
                out.write(adjusted_frame)
                with tracking_image_info:
                    st.write(f'''

Frame number: {frame_number+1} (of {total_frames-1})       

(x-position, y-position, width, height) = ''',x,y,w,h)
                x_position.append(x) 
                y_position.append(y)
                w_position.append(w) 
                h_position.append(h)
            # Display the tracked frame
            with tracking_image_box:
                st.image(Image.fromarray(adjusted_frame), caption='Tracked Frame', use_column_width=True)
            frame_number += 1    
                
        #Plot x and y over time  
        t = np.linspace(1,len(x_position),len(x_position)) - 1
        df = pd.DataFrame({
            'Frame number'     : t,
            'Horizontal pixel' : x_position,
            'Vertical pixel'   : y_position,
            'Box width'        : w_position,
            'Box height'       : h_position
            })
    else:
        df=pd.DataFrame()
    # Release the video capture object
    vcap.release()
    out.release()    
    return df    
       
        
###############################################################################
###############################################################################
###############################################################################        
def part_four(df,file_name):
    st.write("---")
    st.subheader('Part four: RESULTS!')    
    st.write('''

A plot of the horizontal and vertical position of the tracking rectagle (ROI) is displayed in the graph below.   

You can zoom in and hover over specific parts of the graph as desired.   

Note, the x-axis is frame number (not time) and the y-axis is pixel position. You will need to convert these to time and distance in your units of interest.

''')
    st.line_chart(
        df, x="Frame number", y=["Horizontal pixel", "Vertical pixel"], color=["#FF0000", "#0000FF"]  # Optional
        )
    st.write("Please use the two buttons below to download the pixel tracking results as a xlsx file, and a movie of the adjusted video as an avi file.")
    
    
    col_a,col_b = st.columns(2)
    # Function to convert DataFrame to Excel and return a BytesIO object
    def to_excel(df):
        output = BytesIO()
        with pd.ExcelWriter(output, engine='xlsxwriter') as writer:
            df.to_excel(writer, index=False, sheet_name='Sheet1')
            # No need to call writer.save() as it is handled by the context manager
        processed_data = output.getvalue()
        return processed_data
    
    if len(df)>0:
        # Download button
        @st.experimental_fragment
        def download_excel_results():
            st.download_button(
                label="Download Excel Results File",
                data=to_excel(df),
                file_name=f"{file_name}.xlsx",
                mime="application/vnd.ms-excel"
            )
        with col_a:    
            download_excel_results()    

    with open(f'{temp_dir}/{file_name}_ADJUSTED.avi', 'rb') as file:
        video_bytes = file.read()
    @st.experimental_fragment
    def download_video_results(video_bytes):
        st.download_button(
            label="Download Video",
            data=video_bytes,
            file_name=f'{file_name}_ADJUSTED.avi',
            mime="video/avi"
            )
        # video_file.close()  # Close the file
    with col_b:
        # Function to convert file to a download link
        if temp_dir:
            download_video_results(video_bytes)

###############################################################################
###############################################################################
###############################################################################

with tempfile.TemporaryDirectory() as temp_dir:
    # Streamlit app
    st.title('Cell squashing tracker!')
    
    st.write('---')
    # File uploader
    uploaded_file = part_one()
    st.write('---')
    
    # If a file is uploaded
    if uploaded_file is not None:
        #Image adjustment and ROI
        vcap,file_name,total_frames,grayscale, contrast, blur, edges, sharpen_amount, subtract_background, lower_threshold, upper_threshold, binarize, binarize_threshold, x,y,w,h = part_two(uploaded_file)
        # Button to start object tracking
        if st.button('Start Object Tracking'):
            #start tracking  
            df = part_three(vcap,temp_dir,file_name, grayscale, contrast, blur, edges, sharpen_amount, subtract_background, lower_threshold, upper_threshold, binarize, binarize_threshold,x,y,w,h)
            #display/download results
            part_four(df,file_name)
    else:
        st.info("Upload a video file to get started.")