Enabling Low Power Computer Vision on Edge Devices
On Device Neural Networks
Most currently available CV enabled IoT devices such as smart cameras, capture images after a trigger (such as motion sensing) and ship them to the cloud for further processing. This approach can lead to higher power, latency and privacy concerns as compared to processing the images on the edge device. Our solutions allow data extraction using Neural Networks on the edge device to preserve privacy, reduce latency, power and bandwidth.
AI Solutions
We focus on building real world audio and video solutions for targeted use cases with in-house developed compact AI models.
@stevenbell teaches a course in Introductory Digital Engineering at Tufts University and has bought manyUPduino’s. He suggested that the ferrite bead be replaced with a PTC as he was seeing students shorting out their boards leading to burning out the ferrite bead.
If true, this is a pretty worrisome failure as the board is used by makers as well as students for introductory engineering courses, many of whom are novices and would likely end up with a similar failure.
We look into the details of this and the proposed solution in the rest of the blog.
A customer reported that the PLL on the UPduino low-cost Lattice iCE40 FPGA board was intermittent. The blog details the process I used to debug the issue.
This work in part motivated the development of the UPduino 3.0.
