VIAsegura is a library that helps to use artificial intelligence models developed by the Inter-American Development Bank to automatically tag items on the streets. The tags it places are some of those needed to implement the iRAP road safety methodology.
These models require images with the specifications of the iRAP projects. This means that they have been taken every 20 meters along the entire path to be analyzed. In addition, some of the models require images to be taken from the front and others from the side of the car. The models yield 1 result for each model for groups of 5 images or fewer.
So far, 15 models compatible with the iRAP labeling specifications have been developed and are specified in the table below.
| Model Name | Description | Type of Image | Classes |
|---|---|---|---|
| delineation | Adequacy of road lines | Frontal | 2 |
| street_lighting | Presence of street lighting | Frontal | 2 |
| carriageway | Carriageway label for section | Frontal | 2 |
| service_road | Presence of a service road | Frontal | 2 |
| road_condition | Condition of the road surface | Frontal | 3 |
| skid_resistance | Skidding resistance | Frontal | 3 |
| upgrade_cost | Influence surroundings on cost of major works | Frontal | 3 |
| speed_management | Presence of features to reduce operating speed | Frontal | 3 |
| bicycle_facility | Presence of facilities for bicyclists | Frontal | 2 |
| quality_of_curve | How adequate is the curve | Frontal | 2 |
| vehicle_parking | Presence of parking on the road | Frontal | 2 |
| property_access_points | Detects access to properties | Frontal | 2 |
| area_type | Detects if there is an urban or rural area | Lateral | 2 |
| land_use | Describes the use of the land surrounding the road | Lateral | 4 |
| number_of_lanes | The number of lanes detected | Frontal | 5 |
Some of the models can identify all the classes or categories, others can help you sort through the available options.
Some of the features now available are as follows:
- Scoring using the models already developed
- Grouping by groups of 5 images from an image list
- Download models directly into the root of the package
It is advisable to install the library in a Python virtual environment. If you use conda, you can do so with the following command:
conda create -n viasegura python=3.10
conda activate viaseguraSupported versions of Python are 3.9, 3.10, 3.11 and 3.12. For more information on how to install conda, you can visit the official documentation
To install the package you can use the following commands in the terminal
pip install viaseguraTo download the models use this link from the repostitory.
Put the downloaded file on a models folder and decompress it with the following command:
tar -xzvf models.tar.gzThe path where the models are must be models/models_artifacts
Remember to put that path every time you instantiate a model so that you can find the artifacts you need to run them.
In order to make the instance of a model you can use the following commands
from viasegura import ModelLabeler
frontal_labeler = ModelLabeler('frontal') # or 'lateral' You can use either "frontal" or "lateral" tag in order to use the group of models desired (see table above)
Also, you can specify which models to load using the parameter model filter and the name of the models to use, (see the table above):
from viasegura import ModelLabeler
frontal_labeler = ModelLabeler('frontal', model_filter=['delineation', 'street_lighting', 'carriageway']) In addition, you can make it work using the GPU specifying the device where the models are going to run, for example
from viasegura import ModelLabeler
frontal_labeler = ModelLabeler('frontal', device='/device:GPU:0') You can modify the devices used according to the TensorFlow documentation regarding GPU usage.
For a full example of use on this package, you can see this
notebook.
In the notebooks folder you can also find example images to test the execution of the models.
I invite you to visit the manual to understand the scope of the project and how to make a project from scratch using the viasegura models.
This package has been developed by:
Jose Maria Marquez Blanco
Joan Alberto Cerretani
Victor Durand
The distribution of this software is according to the following license