CV Coursework 2025

Note: Due to the large file size, all Input Videos are stored in OneDrive.

Program Running Steps

• Step 1: Navigate to the folder where the Python file is located.

• Step 2: In Terminal, run the entire program by using the command:

  streamlit run Coursework_20412961.py

• Step 3: Upload the single Input Video (using the "Browse files" button)

• Step 4: Manually select the hyperparameter values for detection and tracking

• Step 5: Manually select the body boundary visualization method

• Step 6: Click the "Start to Process" button

• Step 7: Wait until the program running finishes

• Step 8: View the Output Video

• Step 9: Decide whether to download the converted Output Video (using the "Download H.264 Output" button)

• Step 10: Check the output metrics tables and conduct evaluation and analysis

Note: For the first time you run it, you need to download the YOLOv5-Medium model. So please make sure that your network connection is available. If the download fails, must clear the cache and then try again.

Several important Python packages and methods used

• OpenCV (cv2) -- Used for video reading and writing, image pre-processing and drawing visualizations.

  Related used methods: "cv2.VideoCapture", "cv2.cvtColor", "cv2.rectangle", "cv2.putText", "cv2.Canny"

• PyTorch -- Used to load and run the YOLOv5-Medium model, perform inference, and manage GPU memory

  Related used methods: "torch.hub.load", "torch.device", "torch.cuda.empty_cache()"

• Streamlit -- Provides a simple UI for users to upload videos, adjust parameters and display output videos and metrics tables.

  Related used methods: "st.slider", "st.selectbox", "st.file_uploader"

• DeepSORT (deep_sort_realtime) -- Enables person identity tracking across frames

  Related used methods: "DeepSort.update_tracks"

• NumPy -- Assists in numerical calculations and trajectory distance computation

  Related used methods: "np.linalg.norm", "np.mean", "np.std"

• Pandas -- Used to format and display metrics tables in the web interface

  Related used methods: "pd.DataFrame", "st.dataframe"

• Subprocess -- Invokes FFmpeg for video format conversion

  Related used methods: "subprocess.run"

• Other basic libraries -- OS, SYS, Time, Types, Wearnings

Environment configuration and installation

My Python Version is "Python 3.10.13" configured through Anaconda Environment

For specific versions of each used package, please refer to the "requirements.txt" file.

You can install them automatically through the command:

pip install -r requirements.txt

Note: If you have the latest version of the TensorFlow-GPU (even though it will not used) package in your current environment, it is recommended to reconfigure a new Anaconda environment to avoid conflicts.