Copyright © 2018 Bootpin Inc.
Flexible, Multiple Platforms
Our software supports several IP cameras and
runs on x86 platforms as well as ARM SOCs.
Supported accelerators include most Nvidia GPUs
(including Tegra K1) and Xilinx Artix-7 FPGAs.
Windows and Linux are supported. For bare metal
(i.e. no operating system) implementations please
Provided hardware requirements are met, our
software can be embedded into devices or run on
laptops, or desktops or servers in the cloud.
Our software is an ensemble of Deep
Convolutional Neural Networks that provide state
of the art action recognition performance.
In order to provide faster than real time (>30fps)
processing, we leverage hardware accelerators
like GPUs and FPGAs.
Most video surveillance software is based on
object detection and motion detection. Objects
like people and cars are detected and tracked
from frame to frame. Based on the motion of
these objects, an understanding of events
occurring in the video is inferred.
Because the objects are tracked as whole units or
'blobs', nuances of their actions are lost. This is
not a problem for applications that count people
or detect zone intrusions for example. In order to
extract richer information however, more powerful
software is required. Are the people in the video
throwing rocks as they walk around? Did someone
open the trunk of a car? Did a player just kick the
Use our software to find out if events such as
these are happening and respond to them as
soon as they do.
Making video archives searchable usually
requires tagging each video manually and this is a
painstaking exercise. An alternative solution is to
use an object detector to search for objects in
individual frames. An even better solution is to
search for object actions. In other words, query
for the verbs rather than the nouns.
Use our software to build better video search
queries based on verbs and obtain richer results.
More and more devices are getting connected to
the Internet of Things. Some of them require
human input but distance or hardware constraints
prevent traditional forms of input.
If your device falls into this category, our software
can be used to recognize gestures and provide a
Our software is modular and exposes an API that
makes it easy to implement bespoke solutions. In
addition to this, we support several hardware
configurations (see "Flexible, Multiple Platforms"