Pilot AI Labs, Inc.

Pilot AI develops deep-learning based computer vision algorithms whose groundbreaking efficiency allows them to run in real-time on a broad range of hardware solutions. This efficiency enables Pilot AI software to run directly on edge devices, powering applications that either do not have connectivity to the cloud, or for which latency or data security is an issue. Pilot AI supplies these computer vision algorithms and tools to Fortune 500 companies as well as the US Government.


Solutions

Pilot AI develops computationally efficient computer vision algorithms that are capable of running directly on commodity embedded devices. At the core of Pilot AI’s product suite are neural networks that are faster than conventional deep-learning based solutions. Pilot AI’s innovations enable customers to run state-of-the-art computer vision directly on edge devices. The efficiency of Pilot AI’s algorithms also means that edge devices with more compute can process significantly more data than with other commercially available solutions. Pilot AI also maintains a large and growing proprietary dataset used to train models. The time and cost incurred by customers aggregating and annotating their own datasets is significant. In addition to Pilot AI's core innovations that make it possible for customers to run state-of-the-art models at the edge, their dataset allows those models to be tailored to customers’ applications.

Pilot AI provides the following models to their customers:

The Head model

The Head model is a generalized model that will detect people's heads in all environments (e.g. indoors and outdoors, day and night, visible light and infrared). The model has the ability to detect someone’s head regardless of hats.

The Face model

The Face model is a generalized model that will detect people's faces in all environments (e.g. indoors and outdoors, day and night, visible light and infrared). The model will detect the face whether the person is wearing a hat or not, or even glasses.

The Body model

The Body model is a generalized model that will detect people in all environments (e.g. indoors and outdoors, day and night, visible light and infrared). The model will detect someone sitting down, and up to 50% occluded. Camera placement may be overhead looking straight down, elevated (retail setting and most security cameras), or waist height (smart doorbell), with varying level of wide-angle distortion ranging from none to significant fisheye distortion.

The Body Vehicle model

The Body Vehicle model is a generalized model that will detect people and vehicles in all environments (e.g. indoors and outdoors, day and night, visible light and infrared). The model will detect people sitting down, and up to 50% occluded.

The Body Vehicle Attributes model

The Body Vehicle Attributes model detects people and vehicles, and also identifies subclasses of vehicles.

The Head Tracking model

The Head Tracking model detects heads and is able to recognize individual heads, enabling person recognition and more robust person tracking through a frame. The model can recognize a person based on a sample image of their face, and track individual faces, even after periods of prolonged occlusion.

Resources

Pilot AI has the ability to support the end to end data collection, data annotations, data storage, model training, model analysis and model deployment. For more information, please visit Pilot AI's website.