Toll Insight spoke with Karim Ali, Chief Executive Officer at Invision AI.
1. What was the inspiration and mission behind your company, Invision AI?
We want to make sense of the visual world. Since day one, we’ve been focused on democratizing Artificial Intelligence (AI) to solve high-value problems across multiple industries. AI is a wonderful new technology, but conventional approaches require big servers and lots of custom work, making scalable deployments difficult. Invision AI was founded to bring about next generation scalable Artificial Intelligence vision systems.
We empower devices, such as remote cameras and other sensors in the field, to interpret the world around them, without relying on costly hardware or cloud connections. We provide real-world, three-dimensional situational awareness – including detection, geo-localized tracking across sensors and sensor fusion – which significantly reduces custom development. Our platform is designed for applications where speed, cost and privacy are paramount. We are deploying products targeting smart roads, smart cities, and autonomous rail; examples of our most mature products are:
Vehicle Occupancy Detection – We have built the world-leading solution to count passengers in vehicles from the roadside.
Intersection Monitoring – We have the only solution on the market that deploys multiple, collaborative cameras across intersections (city, level-crossing or highway entrances and exits) to monitor speed, vehicle reversing, vehicles traveling in the wrong direction or stopped vehicles.
People counting in public spaces – We have deployed the only solution on the market to count people and track their trajectory over a large space, collaboratively across multiple cameras.
2. Your tolling-relevant product is the Vehicle Occupancy Detection (VOD) system. Please tell us more about it.
We set out three years ago to disrupt the Vehicle Occupancy Detection market with a solution that moves away from manual review to full automation and with precision reaching or exceeding 99%. We wanted to offer a cost-effective product that requires no site-specific tuning and can even be moved from site to site, if desired.
Our VOD system is an easy-to-deploy unit installed on the roadside (or median), that enables high-accuracy and automated counting of vehicle occupants in all weather and light conditions. We have a single-lane as well as a dual-lane unit, which not only reduces deployment cost but also supports high occupancy measurements on non-specialized lanes. All processing is automated and done on-site, preserving privacy while enabling support for low-bandwidth wireless connectivity to collect information via a centralized dashboard.
The system is designed to achieve precision of 97.5% to 99.7% (mobile and fixed), which is significantly higher than rates achieved by either human observation or competing systems.
Through the dashboard, authenticated users can get live occupancy information, export historical statistics and review system performance by accessing pictures of vehicle. Faces and license plates are blurred by default to preserve privacy. Automated alerts can be set based on thresholds to help road operators identify abnormal traffic behavior.
Our VOD unit comes with various options, including a License Plate Recognition (LPR) module, integration with a transponder system, and local area monitoring. The monitoring module can detect hazards such as pedestrians, stopped or reversing vehicles, hazard lights and obstacles to alert operators, to control local light intensity or hazard warning signs.
We see a lot of customers that understand the limitations of using manual methods, such as state police for High Occupancy Vehicle (HOV) enforcement. We aim to achieve a 15x cost reduction compared to the cost of police enforcement, together with a dramatic increase in perceived fairness, since we can precisely detect occupancy in a fully automated and reliable fashion.
3. How have you positioned your VOD system as differentiated from what you see as current market offerings?
We have three game-changing differentiators, which leverage our deep AI and vision expertise.
First, we have been able to dramatically reduce the hardware footprint of our system, while expanding its reach to multiple lanes and enabling either fixed or mobile deployment in under two hours. The system can use standard power with an option for batteries to enable 24/7 operation where power is only available at night. No gantry installation is required. No cable connectivity is required thanks to the use of wireless (LTE) connectivity and ultra-efficient processing taking place directly in the unit.
Second, we have best-in-the-industry accuracy with high reliability. This translates to low operating costs for operators who no longer need to manually review images.
Third, we can address use cases such as hazard detection as added value with the same unit, improving the ROI of deploying our system. We can detect hazard on the road. Soon, we expect to be able to detect, classify and count trucks transporting dangerous goods on the road.
4. You’ve established a specific way to measure VOD performance. What was the driving need and outcome of this effort?
VOD is a relatively new technology in the toll industry and has not yet been widely adopted, partly due to skepticism on performance and reliability. We believe the best way to overcome this resistance is to conduct objective verification of the actual performance of such systems out in the field. Through projects we did with several clients across the globe, e.g., Transurban in North America and Ayalon Highways in Israel, we learned about our customers’ challenges and needs:
The lack of a common language and apple-to-apple comparisons.
The recognition that VOD performance is not only about good AI classification or counting algorithms but also about acquiring high quality images and successfully capturing all vehicle passes.
The transparency and trust gap: road operators can’t rely on each vendor reporting on their performance. They need to be able to independently verify the results.
To address these challenges, we provide online dashboards for our customers to immediately pull all the raw data, meta data, and occupancy predictions. We also provide a set of analysis tools to generate human annotations and then compare the results against the annotated ground truth from randomly sampled natural traffic as well as scheduled controlled tests with hired cars and occupants. Finally, we provide weekly performance reports based on ground-truth generated by independent annotation firms. All the data and code are transparent, open, and accessible. You can learn more about this by visiting our open-source code repository here. It’s with these tools and evidence that we report the industry-leading performance metrics in our joint press releases with our customers!
5. Looking forward, how do you expect that your technology to impact the transportation and tolling industries?
We see our product as an enabler of new policies that will disrupt the transportation and tolling industries. Everybody agrees that we need to find better ways to address congestion, reduce greenhouse gas emission and improve safety, but tools at the disposal of policy makers and road operators have been, to some extent, outdated in our opinion.
Instead of focusing on vehicle level statistics, we now enable visibility into the vehicle and thus enable new policy options not previously possible such as dynamic tolling decisions based on occupant counts.
At the same time, our technology can be deployed cost effectively and can extend to multiple lanes without the need for gantry cameras. This enables road operators to deploy such a system faster, including in places not specially designated as managed lanes and even in urban environments.
Finally, there are many safety improvements that can be implemented using the same technology. For example, it is possible to detect cell phone usage, seat belt violations or dangerous good signs affixed on trucks.
Thank you, Invision AI, for sponsoring Toll Insight!