This project processes rail wagon videos to analyze their status. It determines whether a wagon is loaded or unloaded, calculates the volume of materials if loaded, and detects damages if unloaded. The output includes a report with volume calculations or damage detections along with captured images.
- Uses a custom-trained CNN model to detect the front and rear edges of wagons.
- Counts the number of wagons in the video.
- A trained CNN model classifies each wagon as either "Loaded" or "Unloaded."
- Utilizes DepthAnything Large (Hugging Face) for depth estimation.
- Calculates the volume of materials using depth maps and known wagon dimensions.
- Generates a PDF report with the volume details.
- A custom-trained CNN model detects damages such as cracks or debris.
- Saves images of damaged wagons along with their wagon numbers.
- Generates a PDF report with detected damage details and images.
- OpenCV - For video processing and image manipulation.
- Detectron2 - For object detection (front and rear edges of wagons).
- TensorFlow/Keras - For CNN models (wagon classification and damage detection).
- Torch - For handling deep learning models.
- DepthAnything (Hugging Face) - For depth estimation to calculate volume.
- NumPy & Matplotlib - For numerical operations and visualization.
- ReportLab - For generating PDFs with analysis results.
pip install opencv-python torch torchvision detectron2 tensorflow numpy matplotlib reportlabRun cells in final.ipynb
- Prints and saves a PDF report with volume calculations.
- Detects and saves images of damaged wagons.
- Generates a PDF report with wagon numbers and damage details.
- Gnanendra Naidu N
- Aditya Ranjan
- Bhavya Mashru
- Vikas Sanchaniya
- Priyanshi Shah