This tool is a simple interface that allows injection of image faults into robot cameras. Thanks to this interface, you can create new image libraries by injecting the fault types you have determined, both real-time to ROS cameras, and to the image libraries previously recorded by these cameras. For more information about the purpose of this tool: https://arxiv.org/abs/2108.13803

Developed by Alim Kerem Erdogmus from Inovasyon Muhendislik (c) 2021

ROS Wiki page: http://wiki.ros.org/camfitool

For trying ROS package version of CamFITool (v1.3):

git clone https://github.com/inomuh/camfitool -b noetic-version

Fig 1. Camera Fault Injection Tool Offline FI Configuration

Fig 2. Camera Fault Injection Tool Realtime FI Configuration

• You can apply the faults you choose from the configuration menu to the images in the image library you want and save these wrong images to the folder you want. So you can create your faulty image library.
• You can apply these bugs to all images and also to any number of images, creating a mixed library of bad images without touching the remaining images (offline mode only).
• You can save the configuration of the fault you applied and view the fault plans you saved as you wish.
• You can specify the fault rate to be applied.
• You can inject 6 different types of fault into .bmp images taken from robotic cameras (other image formats are in the testing phase).
• It is possible to choose what percentage of the images in the normal image database to apply fault.
• It is possible to inject two different fault types at desired rates into a normal image database (it is not yet possible to apply more than two faults to the same database).
• For now, three different fault types (Gradient, Dilation, Erosion) can be applied to the real-time stream obtained from ROS camera topics. Other fault types are being tested.
• You can use the ROS Noetic version (noetic version) or the standard version. (ROS version will be updated)
• You can monitor the ROS Camera node.
• You can specify the realtime fault injection frequency rate to be applied.

Fig 3. Camera Fault Injector Tool Realtime FI Applying Demonstration

• First Commit

• Added Real-time FI mode
• Added RGB Camera FI types (Open, Close, Erosion, Dilation, Gradient and Motion-blur). Now this interface supports two camera types and nine fault types.
• Added Robot Cam button. This button provides to show active ROS camera (tried ROS Noetic) screen (User must enter ROS Camera topic name).
• Robot Camera section is active now. User can enter ROS camera topic name for fault injection and enter ROS fault injecting stream frequency value.
• Added Real-time FI plan saving standart. This plan saving is different than Offline FI plan ones.
• Some bug fixes, new error pop-up messages and other things.

• Added Error Log system.
• Fixed randomized fault injection function issue.

• CamFITool's software codes are revised (using Pylint and Prospector).

• Fixed "Known Issues #4: Offline RGB Randomized FI crash" bug
• Fixed "Known Issues #3: Image Formats"
• New image databases added (These images taken from COIL Database)

• CamFITool main codes are revised.
• Two different fault types can now be injected into the same output folder (Max two fault type).
• An fault can now be injected into a certain rate of an image database (Fault Implementation Rate).
• Reset button added (You can reset the interface with this).
• Added a log system to keep the names of the images that were injected and not.
• Some bug fixes.

• The interface has been updated, the unnecessary buttons have been assigned to the upper toolbar.
• Changed the separation of TOF and RGB in camera fault types. All fault types can now be applied to all image types.
• Removed Open, Close and Motionblur fault types.
• The FI Anomaly Detector plugin has been developed. In order to use this plugin from CamFITool, it is enough to download the plugin from its own repository. Thanks to this plugin, it will be possible to detect anomaly in pictures with CamFITool.

• Fault Injection Anomaly Detector (FIAD) plugin has been added. Thanks to this plugin, it is possible to detect errors in images by using the models you have trained with CNN algorithms (Note: Models and images are not shared because they are used within the VALU3S project).
• Help and About sections have been activated [#4].
• Some bug fixes and improvements.

• Offline Type Fault Injection
• TOF Camera FI
• Randomized FI Feature
• Three Fault Types (Salt&Pepper, Gaussian, Poisson)
• RGB Camera FI
• Real-Time Type Fault Injection (only RGB cameras)
• +6 Fault Types (Open, Close, Dilation, Erosion, Gradient and Motionblur)
• Github Wiki page
• ROS Wiki page
• CamFITool Pypi package (This is Alpha version)
• Reset Button
• Mixed Fault Injection
• Fault Types Rework
• ROS Noetic Integration
• More Fault Types
• ROS2 Integration
• Error Log System
• Improved Help Section
• CamFITool ROS Package version

This work is supported by TÜBİTAK Project under grant number 120N803 which conducted by the İnovasyon Mühendislik.

This work is also done by Inovasyon Muhendislik and ESOGU-SRLAB under VALU3S project. This project has received funding from the ECSEL Joint Undertaking (JU) under grant agreement No 876852. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and Austria, Czech Republic, Germany, Ireland, Italy, Portugal, Spain, Sweden, Turkey.

If the code or data help you, please cite the following the papers.

@article{erdogmus2021manipulation,
  title={Manipulation of Camera Sensor Data via Fault Injection for Anomaly Detection Studies in Verification and Validation Activities For AI},
  author={Erdogmus, Alim Kerem and Karaca, Mustafa and others},
  journal={arXiv preprint arXiv:2108.13803},
  year={2021}
}

@research article { ogummf1054761, 
    journal = {Eskişehir Osmangazi Üniversitesi Mühendislik ve Mimarlık Fakültesi Dergisi}, 
    eissn = {2630-5712}, address = {[email protected]}, 
    publisher = {Eskişehir Osmangazi University}, 
    year = {2022}, 
    volume = {30}, 
    number = {3}, 
    pages = {328 - 339}, 
    doi = {10.31796/ogummf.1054761}, 
    title = {DEVELOPMENT OF A FAULT INJECTION TOOL \& DATASET FOR VERIFICATION OF CAMERA BASED PERCEPTION IN ROBOTIC SYSTEMS}, 
    key = {cite}, 
    author = {Yayan, Uğur and Erdoğmuş, Alim Kerem} 
 }

See the LICENSE file for license rights and limitations (Apache-2.0 Licence).