AI enhances road safety
The solution presented consists in the identification of traffic violations, using AI.
The PoC considers two cases: absence of the seat belt and use of mobile phone. The Servei Català de Transit (SCT) provided images for the evidence obtained through its car registration system and vehicle image capture.
That system / control allows the penalty process automation with better / greater reliability and efficiency
Problem or opportunity
In 2019, the Servei Català de Transit (SCT) imposed heavy penalty to 12.800 drivers for using mobile phones while they were driving, that is an 86,9% of the total sanctioning administrative proceedings for human error.
This solution tries on one hand to reduce the accident rate caused by distractions while driving and the absence of the seat belt raising awareness and producing a deterrent effect on drivers The PoC started in order to reduce accidents, increase security / safety and educate drivers. Nowadays the system is working capturing images of 17 points and up to 60 drivers have been penalty
A second need is to help optimize the sanctioning process, giving reliability in the identification of possible infractions, efficiency in detection and subsequent sanction / fine, increasing current capacity freeing up resources on the mechanical parts of violation reviews.
Increases efficiency and productivity, because with the use of the AI solution and the same human resources we could process many more images, today a multiplier factor of 3, but it is expected to reach a multiplying factor above 50 More control, the deployment of more cameras with this service will become a greater extension and control of the territory. That solution opens the possibility of using the same images to detect other road dangers
The service automation the detection by capturing the images of drivers committing traffic violations: use of mobile phone and absence of seat belt through a system of ML and Computer Vision The System is compatible with any camera or manufacturer capable of capturing a semi-zenithally image (from a portico) of a vehicle
The Google Cloud architecture uses AI algorithms based on convolutional neural networks to analyse the photographs. Specifically, it integrates an EfficientDet D4 Deep Convolutional Neural Network (DCNN) architecture that has been trained with more than 80.000 sample images.
This neural network is responsible for detecting the vehicle and the people and assessing whether people are using a mobile phone or not wearing a belt. The decision of whether a person is a driver or not is taken at a later stage, in post processing, where the position of the persona in the vehicle is considered.
Video analysis Systems require significant computing capabilities. In this case, the server where the system resides has an Nvidia GPU card specifically designed for this type of computing.
Google Cloud has been used in the pilot.