R-CNN project

School project for the course "Deep learning" at Jyväskylän ammattikorkeakoulu. I decided to make multi-object detection project. The objective was to make model that could be used to detect cars from livestream CCTV footage. As a side project I made a screen recording software that can be used to record footage from your screen and apply object detection on it.

YOLOv5 IS NOT INCLUDED IN THIS REPOSITORY. You can find it here: https://github.com/ultralytics/yolov5

Models will be automatically downloaded for screen recorder from torch hub!

Technologies used

I mainly used YOLOv5 R-CNN models for this project. It is a multi class multi object detection model, that I used as a baseline for my project. I also used OpenCV for image processing and PySimpleGUI for the GUI.

Installation

Click to expand!

Requirements

• Python 3.9 (Newer versions might work, but I haven't tested them)

Important You have to install CUDA and cuDNN to your computer. I used CUDA 11.8.0 and cuDNN. I cannot guarantee software working without these. Also, it makes software much faster.

To check if you have CUDA and cuDNN installed with pytorch, run this command in your terminal (the script located in main repo):

python test_torch_gpu.py

Installation

1. Clone the repository
2. Optional: Create a virtual environment
3. Install torch separately first. This might not work if CUDA is not installed, also need modern GPU from nvidia.

pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117

4. Install requirements

pip install -r requirements.txt

Usage

Click to expand!

Screen recording software

Screen recording software is located in screen_recorder folder. You can run it with python screen_recorder.py.

Controls:

• You can choose where to save the video. Need to enable "Save video" checkbox to save video.
• Object detection checkbox means, it will run image through a model and show on the screen with bounding boxes all the objects it detects.
• You can choose which model to use. You can choose from small, medium, large or extra-large model.
• Start recording button starts recording.
• Stop recording button stops recording.
• Choose region button lets you choose region you want to record. (Only works for main monitor)
• The slider on bottom left is confidence threshold. You can change it before and while object detection is happening on the recording to change what is the threshold of certainty the model needs to have to show an object.

Object detection - Car counter

To use the car counter run python count_cars/object_tracker.py.