Aerialtronics has announced the release of the company’s new fully integrated computer vision platform with deep learning capabilities. PENSAR is a stand-alone twin sensor platform that utilizes the GPU-accelerated computing power of the NVIDIA Jetson module to allow for real-time video processing and immediate augmented insights.
According to Aerialtronics, it is the first in the market to integrate the new FLIR Boson thermal IR camera into a drone application, and additionally contains a 30x zoom Sony daylight camera. All of this has been integrated inside a small and light-weight platform. Aerialtronics hopes that PENSAR will change the way operators perform aerial tasks in search and rescue, inspections, documenting assets, surveying, mapping and public privacy.
During flight, PENSAR leverages deep learning networks in real-time to detect, recognize, track, classify and annotate certain objects or deviations of interest and conversely, to blur out and mask objects for privacy reasons.
PENSAR allows users to run algorithms that filter the acquired data and output only what’s relevant to them. This eliminates the need for lengthy and costly post-processing procedures, which can massively decrease the turnaround time between aerial data collection and follow-up operations.
Find suppliers of Video Recording, Streaming and Processing Systems for UAS >
PENSAR will have future capabilities that process the filtered data of interest into augmented geo-referenced reports and push notify them in real-time by leveraging existing communications platforms and asset management tools. This way, operators can accelerate the workflows of business operations by preventing fault locations from being overlooked or lost during post-processing, thereby making it easier to relocate objects of interest for the operations team.
The intelligent platform comes with Ubuntu Linux distribution, an open-source operating system that has all the functionalities of a personal computer. This makes it accessible for all kinds of users to make small or big changes to the platform to customize its capabilities for their desired needs. PENSAR is agnostic and compatible with open-source libraries like Google’s TensorFlow. This provides the user with access to deep learning methods and machine vision developers can automate dynamic decision-making systems quickly and easily.
The platform has connectivity via Bluetooth and Wi-Fi and the ability to communicate via UART (Mavlink), CANbus & PWM, which can be utilized as a companion computer to most applications in drones and robotics.
The capabilities of the PENSAR platform include:
- FLIR Boson thermal IR view / Radiometric measurements 1/1 (3:2)
- Day Vision with 30x zoom (HD 1920*1080) 1/1 (16:9)
- Side by side view (vision / IR) (HD 1920*1080) 1/2 (16:9)
- Sensor fusion (IR overlay on vision image stream)
- Dual-stream parallel onboard recording
- Image capturing
- AI-powered object detection, tracking & classification
- Real-time masking
- Optical character recognition (OCR)
- GPU-accelerated computing
- Ability to run unlimited computer vision powered algorithms
Development of custom hand-coded or deep learning algorithms based on the end-user’s requirements is available upon request. Developers can also take advantage of NVIDIA’s JetPack software development kit, allowing them to utilize the libraries and APIs, samples and documentation needed to code and deploy their own algorithms.