Collision Analysis System GUI (collision_gui.py)

This project provides a Tkinter-based Emergency Dashboard for the Car Crash Analysis system. It integrates the backend Roboflow and YOLO inference logic from analysis.py into a user-friendly graphical interface. Features include live camera feed scanning, manual image uploads, visual bounding box overlays, and automated report logging.

---

Python version and virtual environment

Use Python 3.9, 3.10, 3.11, or 3.12 inside a dedicated virtual environment.

• Avoid Python 3.13+ — inference-sdk and some dependency combos are unreliable or fail to install on very new Python releases.

• Use the same interpreter for every command (python, pip, and your IDE) so dependencies match.

  To set up a virtual environment (e.g., using Python 3.9):

  python3.9 -m venv .venv39 
  source .venv39/bin/activate   # Windows: .venv39\Scripts\activate
  python -m pip install --upgrade pip

---

What to install

Install these Python packages into the activated venv. The GUI shares all the dependencies required by the backend analysis.py script:

Package	Role
opencv-python	Process video feeds, draw bounding boxes, and handle image color conversions
ultralytics	YOLO integration for vehicle counting inside crash zones
inference-sdk	Roboflow HTTP inference client to detect crash scenes
python-dotenv	Loads API keys and Model IDs from the .env file automatically
Pillow	Converts OpenCV BGR image arrays into Tkinter-compatible RGB ImageTk objects

One-liner Installation:

python -m pip install opencv-python ultralytics inference-sdk python-dotenv pillow supervision

(Note: Tkinter is included with standard Windows/macOS Python installations. Linux users may need to run sudo apt-get install python3-tk if not already installed).

---

Roboflow configuration (.env)

The GUI relies on a .env file to securely load credentials before launching.

1. Create a .env file in the root directory.

2. Set ROBOFLOW_API_KEY (required).

3. Choose one inference style to define your model:

   • Workflow: ROBOFLOW_WORKSPACE_NAME and ROBOFLOW_WORKFLOW_ID

   • Hosted model: ROBOFLOW_MODEL_ID (e.g., your-project/1)

---

How to run

From the project directory, with the venv activated and .env present:

python collision_gui.py

Using a DJI Tello / Tello EDU (optional)

By default, Start Scan uses your webcam. To use a Tello EDU camera stream instead:

1. Install the Tello Python library:

python -m pip install djitellopy

2. Connect your computer to the drone’s Wi‑Fi:

   • Power on the Tello.
   • On your PC, connect to the SSID that typically starts with TELLO-.

3. Enable the drone stream in your .env:

USE_TELLO=1

4. Launch the GUI and click Start Scan.

If the app can’t connect to the drone stream, it will log an error and fall back to your webcam.

Manual Override (Pilot Mode)

If the drone is behaving unexpectedly during a scan, click Manual Override to take control.

• Requirements: a connected Tello stream (see section above).
• Behavior: keeps the live video running but pauses ML inference while you fly manually.
• Controls (in the control window):
  • W/A/S/D: forward/left/back/right
  • R/F: up/down
  • Q/E: rotate left/right
  • T / L: takeoff / land
  • X: emergency stop (zero RC)
  • I/K/J/O: flips forward/back/left/right

Dashboard Features

• Start Scan: Accesses your local webcam (cv2.VideoCapture(0)) to run a continuous loop, analyzing a new frame every 1 second to prevent API rate-limiting.
• Stop Scan: Safely releases the camera hardware and clears the simulation canvas.
• Upload Image: Pauses any active live feeds and opens a file dialog to analyze a single local image (.jpg, .png).
• View Report: Opens a pop-up window reading from crash_report_log.txt, detailing the history of detected crashes and vehicle counts.
• Status Panel: Dynamically updates to show the total number of vehicles involved in an active crash scene.
• Safety Perimeter Simulation: Runs a drone navigation animation on the right panel and deploys cones around a simulated crash site. Can be started manually via Run Simulation, and also auto-triggers when a crash is detected.

---

Behavior summary

• Unified Canvas Display: Both live drone feeds and uploaded images are routed through a central renderer that automatically resizes frames to fit the Tkinter layout using Image.Resampling.LANCZOS.
• Automated Logging: Whenever a frame analysis returns a "CRASH DETECTED" string, the GUI automatically appends a timestamp, filename, and result message to a local crash_report_log.txt file.
• Decoupled Backend: The GUI imports CrashAnalysisApp from analysis.py. All API requests, YOLO overlap checks, and bounding box math are executed by the backend app before being returned to the GUI for rendering.
• No OneDrive temp-frame churn: Live scan frames are written to the OS temp directory and cleaned up automatically (instead of writing temp_live_frame.jpg into the project folder).

---

Troubleshooting

Issue	What to try
Missing ROBOFLOW_API_KEY	Ensure your .env file is in the same directory as collision_gui.py and correctly formatted.
GUI Freezes / Becomes Unresponsive	Ensure update_scan() is using self.root.after() rather than a while True loop, which blocks Tkinter's main event thread.
Cannot open camera feed	Check system privacy settings to ensure Python has permission to access the webcam, or verify the camera index in cv2.VideoCapture(0).
Tello video doesn’t connect	Confirm you are connected to the drone’s Wi‑Fi (SSID usually TELLO-xxxxxx). Also ensure USE_TELLO=1 and pip install djitellopy in the same venv you run the GUI with.
AttributeError: module 'cv2' has no attribute...	A corrupt OpenCV installation. Run pip uninstall opencv-python followed by pip install opencv-python.
ModuleNotFoundError: torch._higher_order_ops._invoke_quant	Your PyTorch build is too old/incomplete for your ultralytics version. Upgrade CPU wheels: python -m pip install --upgrade --index-url https://download.pytorch.org/whl/cpu torch torchvision