Ray20’s Promising Initial AI Edge Imaging System Results

Thanks to the magic of the iPhone, everyday, high quality imagery is so familiar to us that we often take it for granted. But capturing and analyzing high resolution aerial images, at a speed, frequency and cost not currently available, isn’t familiar (or accessible) to everyone that can benefit from it. We’re determined to change that.

The initial flight tests of Ray20’s technology were incredibly valuable - they gave us significant insight into what’s working, what isn’t and where we need to focus our efforts to fine-tune our device. But our biggest takeaway? The need for continued improvement to our imagery capture and analytical capabilities.

Because Ray20’s device is designed to capture images at an extremely high cadence, the accuracy of the data produced by our deep learning algorithms is dependent on the quality of our images. The higher the resolution, the better (and faster!) our algorithms are able to read, learn and return powerful insight to our customers.

In order to make these improvements, we went back to square one - ground testing. Our first focus was optimizing the camera settings for better capture and adding passive damping to remove distortion. We turned off the auto-exposure, allowing for customised luminosity setting and faster imaging rate, and balanced the profile and color temperature for more “traditional” outdoor exposure.

After multiple simulations on the ground, we took our edge device to the air. Having the capability to quickly acquire and review images remotely, we were able to optimize the camera’s shutter speed, literally “on the fly.” We are excited to share the outcome, as well as some images. Image quality improved significantly - thanks to the new camera settings, we were able to capture higher levels of detail despite weather conditions, image noise and other factors that often impact quality. Additionally, pairing quality improvements with the tweaks to our algorithms resulted in more accurate object predictions which paves the way for valuable insights in near real time and at unrivalled cadence.

Fig. 1: This is an initial example of object detection done on Ray20 imagery and further work is in progress for training the ML algorithm with larger data sets.

But these improvements aren’t just significant for the Ray20 team, or even the larger Airbus organization - we believe this technology can positively impact countless industries on a global scale. High-quality satellite imagery, while incredibly valuable, is also an incredible expense. Large corporations and governments often have the budget to spend, but the cost is impracticable and unattainable for the rest.

With the current domestic and international issues we’re facing, almost all organizations can positively benefit from more frequent aerial detection. Let’s look at two practical examples:

  • So many of us are witnessing the devastation caused by the recent forest fires across the Pacific Northwest. With the unpredictable nature of these fires, Ray20 could provide firefighters with insight into fire paths, their distance from significant landmarks, and knowledge of fire spread to better inform evacuation timing and routes. Once fires are under control, Ray20 could also offer accurate detection of damage throughout communities and forests alike.
  • COVID-19 has created ongoing, unprecedented challenges for our global community. As economies continue to ebb and flow based on infection rates, Ray20 could provide insight into economic recovery. For example, understanding traffic patterns on major highways can be used as an indicator that more of the population is going back to work, which might encourage companies to create more job opportunities. Similarly, understanding how the number and types of cars in the parking lots of a shopping center evolve over periods can help determine business recovery and behaviour, leading to better demand management of business efficiency.

Now more than ever, affordable access to smart commoditized imagery is what our world needs. Single images are no longer valuable on their own; rather, it’s access to an entire dataset. Once that becomes available, deep learning algorithms can run on thousands of images and provide ongoing insight rather than static data points. Ray20 is designed specifically to do this, and we’re determined to shift the market in our direction.

Linda Ge, our Senior Program Manager, put it perfectly - she notes that our team is “committed to continuously improving and rapidly deploying this technology to better serve the communities we are - and hope to be - a part of.” If you are working on similar technology which shares our vision, or are interested in what our imagery can help you gain insight on, we would love to hear from you. Reach out at ray20@airbus-sv.com.

- Ben Wilson