Open-source facial recognition prototype for makers & innovators
In late 2016, our team and a few external partners initiated the creation of an open source Smart Doorbell software/hardware kit. The kit would allow inventors to autonomously manipulate and explore the capabilities of facial and object recognition, a market that was predominantly untapped due to lack of available software. With this vision in sight, our team built a multi-purpose Smart Doorbell maker kit and mobile application that could recognize faces and package logos.
What began as an exploratory open source project, became an even more powerful product than we could have imagined. With limited hardware and technologies, we were able to provide facial recognition software with 85% accuracy at over a 15-foot distance. Our engineers discovered that facial recognition doesn’t rely on color, only the variance in contrast. Through this discovery, we were able to reduce lag and make the program run more efficiently than expected.
To build the Smart Doorbell kit, we mobilized our leading experts in Product Research, Product Strategy, UX Design, Mobile App Development, and Software Support. Project limitations dictated that our team utilize only free of cost, open-source programming options. We placed a high priority on the cost of the kit and ensuring that the Smart Doorbell would appeal to both affluent markets as well as emerging economies.
Built on the Intel® Joule™ platform, the Smart Doorbell project utilized a Linux runtime on the microprocessor to serve a C# service layer to integrate with the open source C++ OpenCV libraries. We trained and tuned the facial and object recognition system so that it could be optimized for the IoT maker community. The facial profiles were trained by transferring the images from mobile devices directly to the C# service over a pub/sub. This allowed the Joule to focus on the heavy lifting of building the classification. The Intel® RealSense™ camera was connected directly to the Joule through a USB hub, and the librealsense Linux SDK was used to capture and analyze the images drawn from the camera in real-time.
We completed the Smart Doorbell project in just under one year. Upon completion, our team and external partners saw the potential for the Smart Doorbell technology well beyond its current feature set; such as providing security to brick-and-mortar businesses. At its core, Smart Doorbell provides a tunable prototype platform for expanding the science and engineering of facial recognition photo telemetry by serving as an exclusive space for inventors to explore, manipulate and leverage facial and object recognition.
