JeVois: Open-Source Quad-Core Computer Vision Platform
I am fond of robots, constantly monitor the Internet in search of interesting solutions. Accidentally found a very interesting project, I hasten to share information with you. The crowdfunding project JeVois is an Opensource machine vision platform : a video sensor and a quad-core processor in a tiny box of 28 cu. see Delivery scheduled for February 2017. This module is capable of transmitting data via USB or serial port. According to the authors of the project, everything should work out of the box. All that is required is to insert a microSD memory card with your program (support for opencv 3.1 and many other algorithms is declared) and connect it to a computer or microcontroller.
JeVois began to develop as an educational project, as the authors understood the urgent need for a similar module of artificial vision when designing in robotics. There are currently no specialized solutions that can work with raspberry pi or arduino out of the box.
')
JeVois is designed to fill this gap by providing a self-sufficient computer vision module that can provide both video transmission and the transmission of the result of the computer vision algorithm through a serial port.
JeVois works as follows: video is captured from the camera, processed on-the-fly through a machine vision algorithm on its own processor, and the results are transmitted via USB to a computer and / or serial to the microcontroller.
On the computer, JeVois is detected as a USB camera. JeVois Smart Camera is an Opensource project, this implies the most extensive personalization and modification options for your needs. The authors of the project declared three standard operating modes:
A smart camera is a full-featured Linux computer on board. It can work by itself, without connecting to a PC or microcontroller. Thanks to the cooling fan, it can operate at full load without overheating, maintaining a constant processor speed of 1.34 GHz.
In the example given in the video, the processing speed of the algorithm is 73 fps on the smart camera processor, that is, it takes 13.68 ms to process one video frame. Also note that the processor is not fully loaded by this algorithm (148.7% of the load, while 400% corresponds to the full load of all 4 cores of the processor). Therefore, additional algorithms can also work in parallel with video analysis.
Examples of recognition of the marker of augmented reality (ArUco), detection and recognition of objects, detection of the road for autonomous driving.
Tracking 120Hz eyes (camera streams at 120 frames per second, processing is performed at 200+ fps)
By April, the developers promise to create a community and finalize the software that will provide advanced functionality.
The JeVois project was partly made possible by research grants from the National Science Foundation and the DARPA agency.
I really hope for the successful development of this platform, it will help to make a qualitative leap in DIY projects.
Give SLAM for robots!
JeVois began to develop as an educational project, as the authors understood the urgent need for a similar module of artificial vision when designing in robotics. There are currently no specialized solutions that can work with raspberry pi or arduino out of the box.
')
JeVois is designed to fill this gap by providing a self-sufficient computer vision module that can provide both video transmission and the transmission of the result of the computer vision algorithm through a serial port.
JeVois works as follows: video is captured from the camera, processed on-the-fly through a machine vision algorithm on its own processor, and the results are transmitted via USB to a computer and / or serial to the microcontroller.
On the computer, JeVois is detected as a USB camera. JeVois Smart Camera is an Opensource project, this implies the most extensive personalization and modification options for your needs. The authors of the project declared three standard operating modes:
- Transfer via USB of video with transfer through the serial port of the video analysis result (for example, coordinates and content of any QR code that has been identified).
- Transfer text information without video.
- Transmission of video marked the most interesting areas for further analysis on a more powerful PC, for example, using neural networks.
Specifications
A smart camera is a full-featured Linux computer on board. It can work by itself, without connecting to a PC or microcontroller. Thanks to the cooling fan, it can operate at full load without overheating, maintaining a constant processor speed of 1.34 GHz.
Software
In the example given in the video, the processing speed of the algorithm is 73 fps on the smart camera processor, that is, it takes 13.68 ms to process one video frame. Also note that the processor is not fully loaded by this algorithm (148.7% of the load, while 400% corresponds to the full load of all 4 cores of the processor). Therefore, additional algorithms can also work in parallel with video analysis.
Examples of recognition of the marker of augmented reality (ArUco), detection and recognition of objects, detection of the road for autonomous driving.
Tracking 120Hz eyes (camera streams at 120 frames per second, processing is performed at 200+ fps)
By April, the developers promise to create a community and finalize the software that will provide advanced functionality.
Comparison of the Raspberry Pi3 model
The JeVois project was partly made possible by research grants from the National Science Foundation and the DARPA agency.
I really hope for the successful development of this platform, it will help to make a qualitative leap in DIY projects.
Give SLAM for robots!
Source: https://habr.com/ru/post/400393/
All Articles