Team 01: SEER

Inspiration

Grant's childhood friend recently passed away in an emotional driving accident. After hearing some distressing news, his friend became very emotional, unable to focus on the road in front of him. We decided to build SEER to prevent future incidents like this from happening and eventually branched out to encompass more directions and aspects of distracted driving as a whole.

What it does

SEER uses advanced AI algorithms to solve problems in tracking distracted driving. By using neural networks to detect emotion, find the direction the driver's head is facing, and detect when the cabin is getting too loud, SEER is able to automatically detect distracted driving, notify the driver using a speaker cue, and send a text message to the teen's parents as well, all features are intended to prevent distracted driving both now and in the future.

How we built it

Because Wi-Fi and a stable Internet connection in the car are not particularly feasible for the existing car owner market, we decided to build SEER as a local inference framework on the Jetson Nano with TensorFlow and TensorRT. The 3 neural networks used and pipelined are: MTCNN, LFA Emotion, and Landmark-based FMP. All app designs were built in Figma.

Challenges we ran into

The hardest challenge by far was getting the 3 neural networks to work in real-time with a webcam on the Jetson Nano. Although the Jetson is already accelerated with Maxwell CUDA cores, it still requires a lot of optimization through TensorRT and C++ acceleration of Python libraries and modules to run even close to real-time.

What we learned

We learned as a team a lot about what it truly means to bring a startup to semi-market-ready from ground up as well as specific technical details such as edge computing optimization and user interface design tools such as Figma. Also, we learned a lot about how what it takes to pursue and make a short video in the modern, bold, and minimalist style many explosive tech companies are seeking to emulate.

What's next for SEER

SEER will continue to evolve and grow with more technical implementation in the prototype that will incorporate both the actual companion data analytics app, data aggregation over time, and effective Bluetooth connectivity on-demand. The concept extends to a full-featured service and product offering in the near future, and we are planning to develop this as soon as we can.