Eyes on the road
Saturday's AI
As part of Saturday's AI, we had to develop an AI project using some of the techniques we have learned during the course. One key aspect was the project had to be a social one.
What have we done?
What is more social than saving lives?
Background
The cause of four out of ten accidents in Spain is drivers getting sleepy. Source: Heraldo
The project
Let's create a model to identify when a driver is getting sleepy while driving and put it in a low-cost device like a Raspberry PI so everybody can install it in their cars.
Machine Learning
Plus
Raspberry PI
Equals to
Less accidents
Video dataset source
Data set of synthetic videos with different types of situations that allow the analysis of multiple driving situations.
Video dataset source
Different cameras allow to take different angles of view and take a multitude of data from the scene.
Video dataset transformation
The dataset has tags that allow categorizing whether the driver is falling asleep. However, these data have been transformed to train predictive models.
Features extraction DLIB
By means of DLIB feature extraction, the most important points of each driver's face were located in the image.
|
|
Data preprocessing
CSV conversion with the values of the previous EAR result for the eye and mouth positions. The variable AWAKE indicates whether the driver is awake or drowsy.
Data analysis
Descriptive analysis of the variables used.
Data analysis
Sampling of the distribution of continuous variables
Model selection and training
The data are separated into sets for supervised training, using the AWAKE binary tag.
Model selection and training
Cross-validation allows to obtain the best selection of hyperparameters in the selected model, optimizing the training result.
Results evaluation
The confusion matrix shows the false positives and negatives that our model detects, this value being quite low compared to the accurate predictions.
Results evaluation
Raspberry Pi Demo
Links
The authors
- David Pérez Vicens @davidpv
- Placido Antonio López Ávila @PlacidoAntonio
- Francisco José García Rico @frangarcia
More info
