OpenCV AI KIT

Posted on 2021-06-29 Edited on 2024-09-30 In Machine Vision , Stereo

Hmmm, today, Vancouver's weather is cooling down a little bit. As mentioned in my previous blog Detectron2 - 2, I received OpenCV AI KIT yesterday. Today, let's try it out.

1. About OpenCV AI KIT

1.1 OpenCV AI KIT on Kick Starter

It looks there are a variety of OpenCV AI kits, from the official websites of its manufacturer Luxonis:

1.2 How does my OpenCV AI KIT look like?

From my picture above, clearly, I've got a set of OAK-D-WiFi at hand.

1.3 OAK-D-WiFi's Capability

From its official website, OAK-D-WiFi is of unbelievably powerful.

Intel® Movidius™ Myriad™ X VPU: accelerates AI algorithms
DepthAI: brings realtime spatial AI
ESP32: enables both WiFi and Bluetooth

Let's take a look at the device from command line:

1 2	➜ ~ lsusb \| grep MyriadX Bus 011 Device 025: ID 03e7:2485 Intel Movidius MyriadX

2. Demonstration

Let's just run depthai_demo.py and see its performance first.

2.1 Some Findings

2.1.1 Image Resolutions

Resultant depth images are of size 640x400
Resultant color images are of size 300x300

➜  results git:(main) ✗ file depth/depth_03847.jpg
depth/depth_03847.jpg: JPEG image data, JFIF standard 1.01, aspect ratio, density 1x1, segment length 16, baseline, precision 8, 640x400, components 3
➜  results git:(main) ✗ file color/color_03847.jpg
color/color_03847.jpg: JPEG image data, JFIF standard 1.01, aspect ratio, density 1x1, segment length 16, baseline, precision 8, 300x300, components 3

2.1.2 More depth images than color images

depthai_demo.py successfully starts inquiring a depth image earlier than a color image, and stops inquiring the depth image later than the color image. In order to use both color image and depth image at the same time, we should make sure both color image and depth image are successfully inquired at the same frame index.

➜  results git:(main) ✗ ls color | wc -l
3736
➜  results git:(main) ✗ ls depth | wc -l
3797

2.2 Results

2.1.1 Single Image

Let's take a look at my demo (yeah, I'm the player), the resultant color image and depth image with index 1000.

color_01000	depth_01000