Advanced Driver Assistance Systems and Intelligent Transport Systems have been a part of our life for a long time already. Computer Vision based components play an important role in creating holistic solutions that cover security and monetization aspects of such systems. We deliver readily embedded products to automotive markets, both primary and after market, and help Smart Cities manage their transport networks efficiently.
Before we enter the driverless vehicle era, we still have a lot of drivers to take care of. Just as the gradual transition from semi-automated driving to fully autonomous vehicles occurs, in-car driver monitoring systems have evolved greatly in the past years, which have improved upon existing functions while also adding numerous new functions. Making our face detection and face attributes tracking technology available for the automotive industry was a natural choice for VisionLabs. Our ability to run state-of-the-art CNN-based algorithms on embedded systems without loss of quality while maintaining high processing speed is our key advantage.
We run the following modules of LUNA SDK in offline mode on most common SoCs:
Head pose estimation
Eye status monitoring
The engine is equally efficient for RGB and IR camera video stream processing and is impartial to face angles, partial occlusions, and accessories like glasses and headwear.
Combining raw data from each module, it is possible to create sophisticated business logic and apply it to various DSM scenarios. For example, drowsiness detection can be implemented as a combination of mouth zone monitoring using the face landmarks tracking module to interpret yawn occurrence and measuring the time during which eyes were continuously closed by utilizing eye status monitoring module.
Adding a face recognition layer to DSM system opens a whole range of new and interesting cases and applications for personal security, human, and asset protection in the private sector and potentially national security. Connecting this functionality to vehicle telematics introduces a new level of data quality and control.
We separate the functions that are run locally on the SoC and face recognition with other useful attributes estimation like age that are run on the server for information security reasons. Storing face templates or any other personal data on in-car modules is both impractical since in this case there should be an interface for sending face template to the device which is also not secure at the same time. Compact size of face best shot allows using secure low-bandwidth communication channel between the vehicle and service provider.
For personal security, the vehicle engine would only start when the allowed driver is behind the wheel, preventing:
- Children from using the car
- Minors in the front passenger seat
- Theft and unwanted use of the vehicle
Besides that, if external view cameras were to be installed, you could unlock the car if the approved operator of the vehicle was recognized while approaching the vehicle.
For the private sector, the following business logic could be implemented:
- Only the appointed driver of the vehicle may start the engine, eliminating unauthorized use
- Record the hours driven by a driver
- In taxi fleets, auto rentals and carsharing service companies, you can make sure that only licensed, authorized, and insured drivers are behind the wheel
- Prevent car theft and inappropriate use
For security, you could prevent illegal seizure of a vehicle, by shutting the engine off remotely and immediately, and the suspect’s face could be captured and matched against a wanted persons list.
The above cases could potentially be beneficial for the following entities:
Fleet management companies
Freight forwarding companies
Municipal transport agencies
Antitheft systems developers
Getting more data out of city traffic monitoring is the next step to enhancing current Intelligent Transport Systems’ performance. Proactive video surveillance systems for vehicle data analysis must operate under various conditions, preferably, on top of existing installations.
Our CNN-based algorithms would return recognition results with the following data:
- Vehicle types
- Vehicle makes
- Public transport types
Together with car license plate recognition, vehicle classification technology enables improved performance of ITS as described in a sample scenario below:
Not only it is essential to implement additional logic such as this when working with the traffic monitoring camera stream, but also it has proven to be efficient in increasing the quality and speed of license plate recognition. Thus, you do not search across the whole database of your registered vehicles, you only search against specific types and makes recognized by the VisionLabs system.
Some cases are known already, but there are plenty to come soon. We cooperate with leading car manufacturers and TIER-1 suppliers in search of applications to help you have a safe and enjoyable ride.
- Emotion recognition to monitor driver and passenger satisfaction, including driverless vehicles
- Personalized experience with infotainment systems and parent control of content display
- Seat belt detection with no possibility to spoof the system with a dummy plug
- Transit advertising on digital screens for specific audience, including offline retargeting