The PSA Group is a French automobile manufacturer, leader in France and second in Europe with 3.5 million vehicles sold in 2019. With a turnover of 74.7 billion euros and more than 209,000 employees worldwide, PSA is a major historical player in the global automotive market. Relying on its production in France, the group wants to ensure freedom of movement by offering safe, sustainable, affordable and enjoyable mobility.
Example of driving video analysis – Detection of nearby vehicles
PSA is taking part in the European research project L3Pilot, which will test autonomous vehicles in 11 European countries for 4 years in real conditions. The aim is to evaluate the technical aspects of autonomous vehicles, user acceptance and their impact on traffic and society. Levels 3 and 4 of autonomy will be tested, i.e. the total autonomy of the vehicles, thanks to sensors and cameras equipped with Computer Vision. This project is part of PSA’s global movement towards vehicle automation.
Following an international call for tenders, PSA began working with Deepomatic on 1 January 2019. The carmaker had several thousand hours of driving time captured by vehicles equipped with cameras, and wanted to set up an image recognition platform in order to equip its operational teams and speed up the classification of videos needed for driving assistance. Thanks to the Deepomatic platform, visual automatons were created capable of processing these thousands of hours of video each month. The automatons fulfill two functions:
1) Blurring the personal data (license plates, pedestrians) contained in the videos in order to comply with the GDPR.
2) Extracting videos of driving scenarios by automatically detecting certain situations (e.g. a vehicle leaving a motorway, whether it is day or night, the presence of a tunnel or hard shoulder). These videos will then be used to test the driving assistance systems they develop in real-life situations.
Automation : The volume of videos cannot be processed manually – it would represent a cost of more than 1M€ / month. The automatons thus make it possible to carry out this work at a much more interesting cost.
Regulations : PSA ensures that it complies with the GDPR regulations (General Regulations on Data Protection). This department eliminates the associated risk which amounts to 4% of PSA’s turnover.
Team autonomy : The objective is to make PSA engineers autonomous to build deep learning models.
IT Architecture : The project is deployed on the cloud in order to integrate Deepomatic applications with the analysis of other types of data, such as speed camera data. Deepomatic applications can also be deployed on local servers and will thus allow massive video anonymization while limiting the dependency and costs of cloud resources.
Collaboration : Deepomatic, PSA, Capgemini and Microsoft teams have fully and effectively collaborated to integrate the solution with PSA’s business teams.
Example of vehicle detection while driving
“The stakes around computer vision are high for the design of our future vehicles. For example, AI contributes to the automatic analysis of our videos from our driving tests and to the respect of regulations (GDPR and autonomous vehicle approval). Deepomatic democratizes the understanding of neural networks and allows us to quickly benefit from the associated gains.”