NEXYAD Artificial Intelligence Technology for Road Safety

 

Customer Possible Hardware
. Smartphone
. Telematics Device
. Dashcam
. Chip
. ADAS Computer
. AD System
. Cloud Server for data receipt

Risk Assessment by NEXYAD IP SafetyNex
Real time, at each moment, permanently (20 times/second)
Local Sensors
GNSS*, Accels*, camera, radar, lidar, etc.
.
Digital Map
Digital Map*                               *Mandatory
.
Risk Factors Detection
. Short range
Interdistance, presence of vulnerables, optical
Visibility (fog, …) narrow drivable aera, etc.
We can extract feature with our own computer
vision modules :
VisiNex, RoadNex, ObstaNex
. Long range
Analysis of e-horizon : complexity of road shape
(curve, …), discontinuity in pathway (intersections),
functions of infrastructure (pedestrians crossing, …),
charateristic of aera (school zone, …).
Information received from infrastructureor from
other cars (V2X): construction aera, traffic jam,
hazard warnings, …

 

Driving-Risk Assessment

SafetyNex

SafetyNex measures in real time, 20 times per second, permanently during driving, the lack of cautious of driving behaviour regarding  driving context (and in particular, road infrastructure complexity and tricks). This lack of cautious is called “driving risk”. It has been validated that alerting a human driver when this driving risk rises too much reduces accident rate by 20%. This driving risk may also be used  in real time to pilot an ACC or an AD system (servo-control of vehicle speed regarding the context in order to keep driving risk below a max accepted value).

VisiNex

VisiNex

Visibility Measurement

Visibility Measurement

VisiNex measures visibility and detects any lack of visibility (fog, rain, snow, sand, smoke, darkness, …). It can be used for hazard warning and as a risk factor for SafetyNex (optional input to take into account visibility in the driving risk assessment).

It also can be used to detect dust on ADAS cameras. VisiNex is also an optional input to ADAS computer vision detection systems: e.g. a detection of obstacles (pedestrians, cars, …) by computer vision may use the characteristics of visibility (provided by SafetyNex to “learn” how the aspect of an object (e.g. a car) is modified with a given lack of visibility (e.g. fog).
It allows to extend detection good performance under bad visibility conditions.

Drivable Surface Detection

Drivable Surface Detection

RoadNex

RoadNex

RoadNex detects drivable surface (also called free space).

ObstaNex

ObstaNex

Objects & Obstacles Detection

Objects & Obstacles Detection

ObstaNex detects objects & obstacles : pedestrians, cars, etc.

 

Those modules run on any OS (Android, iOS, Windows, Linux) and may be recompiled on any chip and architecture. NEXYAD also proposes special optimizations for customers hardware on demand.
Those 4 modules are hybrid Artificial Intelligence: physics, algorithms, knowledge based system, deep learning, neural gas, signal theory, possibility theory, … They run very fast on classical chips.
We are currently working on making them SIL approved.


 

If you want to know more about
our onboard real time Driving-Risk Assessment :

What is Driving Risk ? Anticipation and ADAS Car Insurer Risk & Accident, and UBI
The Vocal Driving Assistant The Smart Robotized Vehicle The Car Manufacturer Director Point of Vue
Rainy Curve Use Case with SafetyNex Tighed Curve Use Case with SafetyNex Priority Use Case with SafetyNex
Stop Sign Use Case with SafetyNex Pedestrian Crossing Use Case with SafetyNex SafetyNex App Onboard Demo

 

Artificial Intelligence and Applied Maths software tools and methods

NEXYAD has been developing a software development environment that is for sure one of the best development environment in the world. It guaranties a high quality level of all our products.

NxMagic: builds automatically a hybrid deep learning & computer vision algorithms based solution from a labelled images database. Testing more than 250,000 feature extractions from images and videos, automatically selecting the N (e.g. N=20) best non correlated features, building several Neural Networks architectures automatically, and training them. Result is the best automatic solution, starting point of our engineering work (gain of time I estimated > 6 month to 1 year on complex pattern recognition tasks).
Note: our deep learning is homemade and beats TensorFlow (see benchmarks in “News”).

NxData: software implementation of the methodology AGENDA (Approche Générale des Etudes Neuronales pour le Développement d’Applications) published in scientific papers by NEXYAD founders. This methodology aims, among other things, to write formal specifications of learning and test databases. Result is much smaller databases with a much performant and robust deep learning result.

NxDeveloper: the “great link” program of our development environment that includes:
. a NEXYAD script language for very fast R&D and tests of leads
. HMI
. program interfaces
. workspace

NxObject: a collection of C++ objects with more than 3,000 algorithms and methods for applied maths, data analysis, signal processing, computer vision, image processing, decision making, artificial intelligence (machine learning and knowledge based systems), used and reused into dozens of projects then fully debugged.

NxAtelier: extract of NxObject specialized in visibility measurement. Connected to a physical test bench system used to calibrate cameras. After calibration VisiNex (visibility measurement) can run on any decent automotive camera and lead to exactly the same results.

NxNeural: it is a part of NxObjects interfaced within Microsoft Excel to apply our deep learning, and data analytics methods without programming (useful for Data Scientists)

Encapsulation: of our real time AI module SafetyNex into applications environments: Android Java, iOS Objective C, Linux C, Windows C, to simplify work of integrators.

Asynchronous real time replay architecture with: MAP data, timestamped sensors data and images, … using the framework RTMaps (proposed by the company INTEMPORA). This complete asynchronous real time replay environment is used to recreate bugs encountered by customers, analyse and understand them to get them fixed quickly.

This unique environment is a key element that generates deserved trust of our customers: we are over state of the art in Artificial Intelligence, and obsessed with source code quality, validation, methodologies: if something works, we know why and how it does. We are also collaborating with specialized companies in SIL (Security Integration Level). And we are agile to develop new real time modules in very short times when needed, always up to date.