深度学习能在很大程度上提高计算机视觉系统的准确性，从而提供革命性的应用，如智能安全摄像头和自动驾驶汽车。但是，要提高计算机视觉系统的准确性，对计算机性能的要求极高，从而限制了这些应用所带来的裨益。而PilotAILabs的使命就是通过降低这些算法对计算性能的需求，来推广这些革命性的人工智能应用。今年，PilotAILabs的软件将被部署到数百万台设备中。

Pilot AI Labs- A deep-learning system for embedded devices. Developed technology that allows computers in security systems and drones to track gestures and people, focusing on specific objects, and estimating distance.

Using Pilot AI Labs proprietary deep learning algorithms, we demonstrate robust finger tracking and gesture recognition using the camera from a standard Nexus 6. No additional depth sensors or other devices were used.

Using our real-time deep learning technology, we demonstrate the ability to follow any object using only the input from a camera. The video shows the real-time camera feed that the drone processes (bottom left) to follow targets of interest, including two different people and a car.

Using the same deep learning algorithms, we demonstrate ‘follow-me’ for people doing different activities. The drone is able to follow a soccer player as he makes quick movements and track a target even as he is occluded behind another person.

Using Pilot AI Labs proprietary deep learning algorithms, we demonstrate depth estimation of an object using only a single USB webcam. No additional depth sensors or other devices were used.

Using Pilot AI Labs proprietary deep learning algorithms, we demonstrate localization of the camera relative to another object using only a single USB webcam. No additional depth sensors or other devices were used.

Using a 5.8 Ghz analog video link, we stream video from the GoPro attached to the drone down to a small computation device (Odroid XU4, equivalent to a Samsung Exynos), apply our state-of-the-art deep learning algorithm to detect and track the target, and transmit controls back to the drone using the wireless telemetry transmitter.

Using the Intel RealSense, our algorithms are able to robustly detect and track objects even at night, with extremely little light. Our deep-learning based approach intelligently combines all the information provided by the RealSense, enabling computer-vision in this scenario where a traditional camera would fail. Combined with our computer-vision based drone control, our follow-me technology works in both daytime and nighttime, without a GPS tracker.