-
Purpose of work: Abnormality Detection using 2D picking model
-
Detailed description: Implementing a python script (test_abnormality_detection.py) that receives as input arguments:
- the trained 2D picking model path
- the objects' image that is subject for evaluation (normally a camera shot showing objects in the working scene)
- the allowed regions' image, which is used as reference map for identifying the prohibited and allowed zones for the objects; this map is a B&W image where white pixels indicate the allowed regions on the scene
-
Destined output:
- returning the judge_image results as output: a list of dictionaries that satisfy the input threshold, where each dictionary refers to an object detected by the 2D picking model (call of grasping_inference inside judge_image), with their visualization image;
main object details:
- coordinates: 'x' and 'y'
- orientation angle: 'beta'
- confidence rate of model's object detection: 'confidence'
- judgement of the position: 'inside_allowed_region' (in boolean value)
- saving the visualization as an Image file on the same folder as the input, under the format "inputImagePath_timestamp.jpg"
- printing the list of objects that are outside the allowed region
- plotting the detected items' centers on the map with colorcoding (in blue if inside allowed region, else in red)
- returning the judge_image results as output: a list of dictionaries that satisfy the input threshold, where each dictionary refers to an object detected by the 2D picking model (call of grasping_inference inside judge_image), with their visualization image;
main object details:
-
Necessary Presets:
- Data acquisition: model and sample image files are available on XC's NAS server:
- Grasping Model: \192.168.100.95\pub\A4_Kowa-Optronics\6_2D_picking\mlserver_test\kowa_infer.zip
- PS: The user needs to unzip the file, and the model path should point to the "grasping" folder
- PyTorchGrasping submodule:
- initialize the submodule after cloning the project:
git submodule update --init --recursive
- install https://github.com/xc-jp/PyTorchGrasping/tree/kowa/v1.2.1 requirements
pip install -r .\PyTorchGrasping\requirements.txt
- set PythonPath for including the submodule on the interpreter:
- on Windows DOS:
set PYTHONPATH=.;.\PyTorchGrasping;
- on Powershell:
$env:PYTHONPATH+='.;.\PyTorchGrasping'
- on Ubuntu or Mac terminal:
export PYTHONPATH=.:./PyTorchGrasping:
- on Windows DOS:
- initialize the submodule after cloning the project:
- Data acquisition: model and sample image files are available on XC's NAS server:
-
Running the script:
- run the script as the following examples:
- No GPU:
python .\test_abnormality_detection.py <path\to\>\kowa_infer\grasping .\samples\test_images\00000001_clear_center.jpg .\samples\allowed_regions.png --conf_threshold <value>
- With GPU (you must have cuda installed on your environment):
python .\test_abnormality_detection.py <path\to\>\kowa_infer\grasping .\samples\test_images\00000001_clear_center.jpg .\samples\allowed_regions.png --gpu --conf_threshold <value>
- No GPU:
- run the script as the following examples:
-
WORK STILL UNDER CONSTRUCTION:
- judging the angle of inclination for each object
- judging the relative distance between objects
-
Notifications
You must be signed in to change notification settings - Fork 0
xc-jp/kowa-Robot_ABD
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
Anomaly Behavior Detection focused on the movement of AGVs.
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published