Detects cars in parking lots, inventory on shelves, etc., and counts vacant information.
- Empty area detection
- Export count to csv
- Windows, macOS, Linux
- Python 3.7 and later
- ailia SDK 1.2.10 and later
- torch for grid sampler (
pip3 install torch
), torch is not required if you use ailia SDK 1.2.14 and later
- Put this command to open GUI.
python3 ailia-apps-empty-detection.py
- Push "Input video" button to select input video
- Push "Set area" button to set area
Click on the screen to draw two lines.
- Push "Run" button to execute the app
a. Push "Output video" button to select the path for video b. Push "Output csv" button to select the path for csv c. Push "Output image" button to select the folder for detected image
The examples of csv file.
sec , time , area0 , area1 , area2
0 , 2023-02-18 12:32:52.289135 , 1 , 1 , 0
You can improve performance by lowering the image recognition resolution.
Specify the ratio of whether or not to judge that an object exists when how much it overlaps.
By default, all class labels are recognized. By entering the class name in Accept Label, only a specific class label can be recognized. For example, in the case of a parking lot, area determination can be performed only for the car by setting the Accept Label to car. Multiple labels can be specified by separating them with commas.
By default, all class labels are recognized. By entering the class name in Deny Label, ignoring certain labels. For example, by specifying person, it is possible to remove the effect when a person crosses the camera. By default (none), all objects are detected and not ignored.
By default, 1 object is assigned to 1 area that overlaps the most. Enable Multiple Assign to assign one object to multiple areas.
SwinB is a highly accurate model using VisionTransformer. R50 is a traditional model using Convolution.
Lvis can detects 1000 objects. In2k can detects 21000 objects.
classDiagram
`ailia APPS Empty Detection` <|-- `Detic` : Empty area detection (area matching)
`Detic` <|-- `ailia.core` : Inference result
`ailia APPS Empty Detection` : Input Video, Output Video, Output csv
`Detic` : Large object segmentation
`ailia.core` <|-- `onnx` : Model
`ailia.core` : ailiaCreate, ailiaPredict, ailiaDestroy
`ailia.core` <|-- Backend : Acceleration
`onnx` : detic
`Backend` : CPU, GPU
https://pixabay.com/videos/building-parking-lot-parking-car-130571/