Project about Real-time Object Detection and Localization, for the Computer Vision and Image Processing course at LTU.
The Python code of the project is located in the src/ folder.
The resources used in the process, which are mostly templates, are located in the res/ folder, along with the deep learning IA models.
The dataset images to test the process on should be located in the dataset/ folder.
Images used in this project are not uploaded to the github repository.
You should first download them here in the Project_1 folder.
You should then place the Project_1 folder's content in the dataset/ folder of the project, such that it contains the raw and rosbags folders.
No training dataset is provided in the project. We used the dataset which can be found here to train our models.
The trained models are located in the res/train_models folder.
To run the program, you simply have to run the res/main.py Python script.
If you wish to modify parameters, please do so in this main.py file.
Parameters that can easily be modified are part of the #? XXXXX (can be modified) ------------- blocks of code.
Before going to the main process loop, the program first shows the base template that will be used for the matching step of the process.
Press any key to continue to the main processing loop, or [Escape] to cancel the process.
During the processing, press [Space] to switch from the image per image mode to the continuous mode.
- During the image per image mode, press any key to go to the next image.
- During the continuous mode, the processing keeps going as long as it can
During any of these mods, press
[Escape]to stop the program early and quit.
Here is an example of the result of running the program with the default processing modes.
They are the AI mode for the pre-processing step and the Background scoring method to detect fake and true extinguishers.
The outer boxes show where extinguishers have been detected at the end of the pre-processing step. The inner boxes show where the main body of each of them has been detected (used to decide which depth value to use to locate it). The boxes' color show whether they are considered fake or not.