In the last blog article, we took a quick look at the concepts of AI and machine learning technology. In this post, we will explain one of the biggest differentiators of Wyze’s AI technology — Edge AI — and why it’s important to Wyze.
What is Edge AI?
Edge AI is a branch of Edge computing. Edge computing is an emerging technology aimed to address cloud computing’s current shortcomings. Edge computing brings computation and data storage closer to the location where it is needed, to improve response times and save bandwidth. In the AI field, this means running inference (a trained model) to process data collected by sensors directly. This mechanism processes data locally, instead of sending data to the cloud, and is therefore independent of the internet connection. It’s widely used in time-sensitive areas, such as robot gesture control, self-driving cars and real-time object detection.
Why did Wyze choose Edge AI?
We looked at many factors when making this decision. Since Edge AI doesn’t utilize network traffic it has three major benefits:
- Shorter response time.
The computation starts with the event recording. The model extracts frames from the video stream and computes the confidence level of a “person” in each frame. This saves the time it would otherwise take to send data to the cloud and then run the inference on the cloud. As a result, the camera knows the existence of a person in videos immediately.
2. Lower cost.
Cloud storage and computing costs can be significant and they grow as the number of users grows. By removing the cloud component and thereby removing the cloud costs, Wyze can provide this technology to its existing users without charging additional costs like a monthly subscription.
3. Better privacy and security.
Computing on the Edge doesn’t require sending the video to a cloud AI computing machine for processing. The only communication with the cloud is sending the original video for 14-day cloud storage. Reducing the amount of data sent to the cloud allows for more efficient and secure transmission.
How did this happen?
Wyze partnered with another Seattle-area startup called Xnor.ai to bring this cutting edge technology to our users. Check out the story from Xnor.ai about how we made it happen.
What’s next?
Although there are many advantages to employing Edge AI, there are also challenges as we look ahead. The first is that like many other speed-resource trade-offs, we have the challenge of how to make the inference more accurate given the limited hardware resources available in our cameras. Another is dealing with power consumption. So far the inference is run on Wyze Cam and Wyze Cam Pan, which are plugged into power. What if we want to run it on a battery-powered device? Or on a device with a less powerful chip? In addition, we may want to run other detection models or employ other AI technologies that could expand the capabilities of our cameras even further.