Classifying and measuring vehicles

This project has been developed at the Kempten University of Applied Sciences for the lecture "Bildverarbeitung und maschinelles Sehen". All rights reserved

The Team

• Jakob Bleickert: Video processing
• Lukas Harzheim: Taking measurements
• Konstantin Paulus: Image classification

Project Structure

/saved_models
In this directory you can find the ML model that has been exported from google colab

/presentation_materials
Here you can find meterials that have been used for the presentation

/notebooks
This directory has been used in order to develop the algorithms

/datasets
Here you can find the videos and images that have been used for the image processing tasks

Important Notebooks (/notebooks/*)

video_preprocessing.ipynb
Is extracting a frame, containing a vehicle in the center of the image, from a video sequence

take_measurements.ipynb
Is measuring the vehicle inside a given frame

classification_model_training.ipynb
Can be used in order to train the classification model (High GPU and RAM requirements)

classification_daihatsu.ipynb
Is applying the model to one of our datasets

classification_rule_based.ipynb
Tries to evaluate an alternative approach for classifying the image

combination_of_algorithms.ipynb
Is combining all algorithms that can be used to process a given video

Testing the source code

*The .ipynb files can be executed with jupyter notebook
Make sure you have the following packages installed in your environment:

# Requires the latest pip
$ pip install --upgrade pip

# Current stable release of tensorflow for CPU and GPU 
$ pip install tensorflow==2.5.0

# For processing images
$ pip install opencv-python==4.5.2.54

# Displaying charts and images
$ pip install matplotlib==3.4.2

# For scientific computing
$ pip install numpy==1.20.3

# Might come in handy
$ pip install scipy==1.7.0

Keywords

Machine learning, computer vision, deep learning, CNN, vehicle classification