Analyzed a dataset consisting of videos recorded at a bus entrance in China and annotated the images in video frames. Trained a model using TensorFlow Object Detection to count the number of people passing through a specified scene.
The videos were annotated frame by frame, each second, and then annotated and split using Roboflow. Yolov5 was utilized to detect people in each image (frame). Below, I will explain the process in detail.
To begin, navigate to the main directory of the project in Google Colab and clone Yolov5. Next, add the torch and os libraries. Then, import the dataset from Roboflow.
# clone YOLOv5
!git clone https://github.com/ultralytics/yolov5 # clone repo
%cd yolov5
%pip install -qr requirements.txt # install dependencies
%pip install -q roboflowAfterward, we need to add the torch and os libraries to our project. This can be done by including the following lines of code
import torch
import os
from IPython.display import Image, clear_output # to display images
print(f"Setup complete. Using torch {torch.__version__} ({torch.cuda.get_device_properties(0).name if torch.cuda.is_available() else 'CPU'})")Once the libraries are imported, we can proceed to import our dataset from Roboflow. This step is crucial for accessing the annotated images and frames. The specific code for importing the dataset may vary depending on the format and structure of the dataset provided by Roboflow.
from roboflow import Roboflow
rf = Roboflow(model_format="yolov5", notebook="ultralytics")os.environ["DATASET_DIRECTORY"] = "/content/drive/MyDrive/AI_Project"from roboflow import Roboflow
rf = Roboflow(api_key="6MZs0IEOrn1yaUepdoxH")
project = rf.workspace("ca-foscari-university-of-venice").project("countpeople")
dataset = project.version(14).download("yolov5")To detect humans, use the following code, which considers the number of epochs, number of batches, and image size.
!python train.py --img 416 --batch 16 --epochs 25 --data {'/content/drive/MyDrive/AI_Project/CountPeople-14'}/data.yaml --weights yolov5s.pt --cacheTo classify objects as humans, set the --conf parameter to 0.7, indicating that if the probability is higher than 70%, it should be classified as a human.
!python /content/drive/MyDrive/AI_Project/yolov5/detect.py --weights /content/drive/MyDrive/AI_Project/yolov5/runs/train/exp/weights/best.pt --img 416 --conf 0.7 --source {'/content/drive/MyDrive/Pellilo/Main/CountPeople-14'}/test/imagesTo read all the images from the directory, you can use the following code snippet
import cv2 as cv
import glob
import torch
import pandas as pd
import numpy as np
data = []
model = torch.hub.load('/content/drive/MyDrive/AI_Project/yolov5','custom', path='/content/drive/MyDrive/AI_Project/yolov5/runs/train/exp/weights/best.pt', source='local')
files = glob.glob("/content/drive/MyDrive/AI_Project/yolov5/runs/detect/exp/*.jpg")
for myFile in files:
img = cv.imread(myFile)
results = model(img)
#results.print()
display(Image(filename=myFile))
data.append(results)
df = pd.DataFrame(data)df.to_csv("df.csv", index=False)# Load the csv file into a pandas dataframe
df = pd.read_csv('df.csv')
# Split the strings in each cell into a list of strings
split_strings = df[df.columns[0]].apply(lambda x: x.split(" "))
# Convert the list of strings into a dataframe
result = pd.DataFrame(split_strings.tolist(), columns=[f"Column {i}" for i in range(1, len(max(split_strings, key=len)) + 1)])result.columns = ['Column_1', 'Column_2','Column_3', 'Column_4','Column_5', 'Column_6','Column_7', 'Column_8','Column_9', 'Column_10','Column_11', 'Column_12','Column_13', 'Column_14','Column_15', 'Column_16','Column_17', 'Column_18','Column_19']result['Column_4'] = pd.to_numeric(result['Column_4'], errors='coerce')
People = result['Column_4'].sum()
print("Total number of peoples in the all images is:", People)