I'm always on the lookout for integrated camera/processor boards that can be used for Vision AI (object detection, face recognition, etc). I recently did a forum post "Has anyone used a person-sensor from Usefulsensors?" about a $10 sensor from a new startup founded by Pete Warden of TinyML fame. Interesting product, but the face detection/face recognition AI model is not updateable (on purpose). I think that I could find use cases for it, but in general I need a sensor where I can update the AI model.
I've just started using the Grove Vision AI Module from Seeed Studios. It is a pretty capable piece of hardware for $25.99.
Specification
| Parameters | Description | Note |
|---|---|---|
| Connector | Grove (Grove base for Arduino) | 5V Charge and Data Transmission |
| Double row 7pin socket (seeed duino XIAO) | 5V Charge and Data Transmission | |
| USB Type-C | 5V Charge and Firmware burn | |
| Communication Mode | IIC | |
| Processor | Himax HX6537-A | 400Mhz DSP (ultra low power consumption) |
| Camera Sensor | OV2640 chip | Resolution Ratio 1600*1200 |
| Microphone | MSM261D3526H1CPM chip | -26dBFs sensitivity |
| Accelerometer Sensor | STLSM6DS3TR-C Sensor | 3D accelerometer and 3D gyroscope |

It uses the HiMax HX6537-A DSP chip as its AI processor. This is the same part that is used on the Himax WE-I Plus (WiseEye) development board that sells for $69.95. That board uses a lower resolution HM01B0 camera (320x240), a 3axis accelerometer, and dual microphones. One drawback of the board as with many newly released Seeed products is that there is a dearth of documentation. I'm hoping that will change over time and I saw on the Discord server that they are working on Edge Impulse support. The WE-I Plus is supported by Edge Impulse, so hopefully that will be a simple port. Another drawback is that the DSP chip is using Synopsis ARC IP which requires setting up a new (to me) toolchain to program the firmware (this is where lack of an API is problematic). I haven't gotten any response to questions on the Seeed Support Forum, so I'll need to try the Discord server.
One interesting aspect of using the board is that all the inferencing information goes out the Grove I2C port and the USB-C is only used for programming and debug. That could provide data privacy in use if there were no physical access to the board (to comply with Sensor 2.0 proposed by Pete Warden, etal).
Another interesting feature is that the board headers are pin compatible with the Seeed Xiao line of boards - so you could stack it on top of one to provide an application interface and communication (Xiao's come in nrf52 and ESP32 versions).
I thought I would do a quick video showing the out-of-box person(face) detection model before trying to build my own models. I've used this board with a Wio Terminal and the SenseCraft application, but for this demo I'm using a Xiao(SamD21) on an Xpansion board. Normally the inferencing info would only be printed to the Xpansion board OLED display and this would be battery powered, but for this demo I have USB connected to the Xiao (for power and Serial Monitor output) and a separate USB connected to the Vision AI board to display the camera image with the bounding box and confidence score.
-
BigG
-
Cancel
-
Vote Up
0
Vote Down
-
-
Sign in to reply
-
More
-
Cancel
-
shabaz
in reply to BigG
-
Cancel
-
Vote Up
0
Vote Down
-
-
Sign in to reply
-
More
-
Cancel
Comment-
shabaz
in reply to BigG
-
Cancel
-
Vote Up
0
Vote Down
-
-
Sign in to reply
-
More
-
Cancel
Children