Broad Range Applications of Real Time Driving Risk Assessment

Posted on

BROAD RANGE APPLICATIONS OF REAL TIME DRIVING RISK ASSESSMENT
Driving risk is not predictable from the so called “black spots” location, or from only driving behaviour. Driving risk appears when driving behaviour is not adapted to driving context, and particular, to road infrastructure complexity. There is no inherently bad driving behaviour (it depends on WHERE you drive: a disused airport ? in front of a school ? approaching an intersection ? risk is different for all those case). There is no inherently dangerous infrastructure and all automotive projects that record “black spots” are doomed to failure : they are places where few drivers in the past had a driving behaviour that was not appropriate to infrastructure complexity, and they died in accident. Thousands, millions, of other drivers did not have any accident at this location. What will this information bring to YOU ? Nothing ! It is necessary to evaluate adequation of YOUR driving behaviour to infrastructure complexity.
An AI module does that 20 times per second: SafetyNex.
Driving risk computed by SafetyNex is a core notion with lots of different applications : car insurance, fleet management, commerce, ADAS, Autonomous Driving, Vocal Driving Assistants, …

SafetyNex Presentation Page

Deep Learning for Onboard Applications: Hidden Trap

Posted on

DEEP LEARNING FOR ONBOARD APPLICATIONS: HIDDEN TRAP
Now Deep Learning is used in onboard detection and pattern recognition applications. NEXYAD for instance uses Deep Learning in RoadNex (road detection without need of markings + detection of free space), and ObstaNex (obstacles detection).
But if you do not analyse your INDUSTRIAL project in detail, you may have bad surprises : everyone thinks he/she knows that the more numerous the training examples, the most accurate the KPIs. Let’s say you used 1 billion km to train and validate your Neural Network (NN) for computer vision. Now a new cam is launched on the market (32 bits per color, 10k) : If you want to use your NN, you will degrade quality of images and put them into your system. If you want to take advantage of your better camera, then you must capture 1 NEW billion km with the new cam and train a new NN.
NOT VERY INDUSTRIAL!
NEXYAD has developed a methodology to get same KPIs with a very picky compact database (easy to reshape the database with new sensors) : A.G.E.N.D.A. (Approche Générale des Etudes Neuronales pour le Développement d’Applications), published in scientific papers in the 90’s – yes – the 90’s by NEXYAD team.

Deep Learning Database Validation

Nexyad at Autonomous Vehicle World Expo 2018

Posted on

AVWE 2018
 
NEXYAD will be present at Autonomous Vehicle World Expo 2018 in Stuttgart, June 5 – 6 – 7
Come to visit Nexyad on Groupement ADAS booth 2015/hall C
Discover our 4 Artificial Intelligence Algorythms for ADAS, Autonomous Driving and Telematics

SafetyNex : Real Time Driving Risk Assessment
Fusion of Digital Map with Sensors, combinable with RoadNex, ObstaNex and VisiNex
Artificial Intelligence giving Safety to your Driver Assistant or your Autonomous Driving Systems

RoadNex : Detection of Road Edges and Free Space on front of Vehicle
Artificial Mono Vision Algorythm for Autonomous Driving & ADAS

ObstaNex : Detection of Obstacles on front of Vehicle
Artificial Mono Vision Algorythm for Autonomous Driving & ADAS

VisiNex : Measurement and Score of Visibility
Detection of Lack of Visibility, Fog, Heavy Rains, etc… on front of Vehicle
Artificial Mono Vision Algorythm for Autonomous Driving & ADAS