Table of Contents
- Introduction
- Getting Started
- Edge Impulse
- Improving Edge Impulse Model
- Testing The Machine Learning Model With OpenMV
- Adding The Water Sprayer System
- Testing The Water Sprayer System
- IoT Ambient Monitoring System | Part1
- IoT Ambient Monitoring System | Part2
- Summary
**********************************************************************************************************************
In this chapter I will show you the procedure to use OpenMV when we are going to test the machine learning model created with Edge Impulse on our Nicla Vision board.
Build the Firmware
Since the Nicla Vision doesn't have any on-board SRAM we need to build the machine learning model into the firmware and load it from the flash. To do so, go to https://github.com/openmv/openmv and fork the repository.
Rename the machine learning model and the label file downloaded in Edge Impulse, in my case I renamed to bee_or_spider.tflite and bee_or_spider.txt respectively.
In your fork, replace the built-in machine learning model under src/lib/libtf/models with the model you downloaded from Edge Impulse Studio. Commit the files and push the commit to the repository. It will build a new firmware automatically.
You can inspect the build process under "Actions". Once the firmware for NICLAV, has been built you can download it from the firmware link.
Flash the Firmware
We can now return to OpenMV and flash the new firmware to the Nicla Vision.
Put the Nicla Vision in bootloader mode by double-clicking the reset button – the green LED will start flashing. Click the Connect button in the IDE – the dialogue to Load a firmware shown below will open.
Click OK, and navigate to bin file produced in the previous step and click Run.
The Nicla Vision will be flashed with the new firmware, which includes the Edge Impulse model.
In my case I uploaded my best machine learning model, which is 1.81 MB in size. In other words, I am using 90.5% of the flash memory of the Nicla Vision.
Run the Script
The next step is to write a Python script in OpenMV to control the Nicla camera and use the ML library to classify the image stream and try to detect our target objects. The video stream is just a series of image frames which are passed to a TensorFlow object which classifies the frame using the model and calculates a confidence prediction. The complete script of the classification example is as follows:
# AUTHOR: GUILLERMO PEREZ GUILLEN
import sensor, image, time, os, tf, pyb
redLED = pyb.LED(1) # built-in red LED
greenLED = pyb.LED(2) # built-in green LED
sensor.reset() # Reset and initialize the sensor.
sensor.set_pixformat(sensor.RGB565) # Set pixel format to RGB565 (or GRAYSCALE)
sensor.set_framesize(sensor.QVGA) # Set frame size to QVGA (320x240)
sensor.set_vflip(True)
sensor.set_hmirror(True)
sensor.set_windowing((240, 240)) # Set 240x240 window.
sensor.skip_frames(time=2000) # Let the camera adjust.
labels, net = tf.load_builtin_model('bee_or_spider_v2')
found = False
def flashLED(led): # Indicate with LED when target is detected
found = True
led.on()
pyb.delay(2000)
led.off()
found = False
clock = time.clock()
while not found:
clock.tick()
img = sensor.snapshot()
for obj in tf.classify(net, img, min_scale=1.0, scale_mul=0.8, x_overlap=0.5, y_overlap=0.5):
print("**********nPredictions at [x=%d,y=%d,w=%d,h=%d]" % obj.rect())
img.draw_rectangle(obj.rect())
predictions_list = list(zip(labels, obj.output()))
for i in range(len(predictions_list)):
confidence = predictions_list[i][1]
label = predictions_list[i][0]
print("%s = %f" % (label, confidence))
if confidence > 0.8:
if label == "bee":
print("It's a BEE")
img.draw_string(5, 12, label)
flashLED(greenLED)
if label == "spider":
print("It's a SPIDER")
img.draw_string(5, 12, label)
flashLED(redLED)
print(clock.fps(), "fps")
How does it work?
- The bee, spider and unknow prediction scores are printed through the serial port;
- The conditional to activate the detection of a bee or a spider must be greater than 0.8;
- If the camera detects a bee, then the green LED lights for 2 seconds and prints the message "It's a BEE" on the serial port; and
- If the camera detects a spider, then the red LED lights for 2 seconds and prints the message "It's a SPIDER" on the serial port.
Test
Below I show you an image capture when the camera has detected a bee.
Also, below I show you an image capture when the camera has detected a spider.
Below I show you the tests carried out with my model created with Edge Impulse and OpenMV. As you can see, I did the tests with images of a bee and a spider printed on a cardboard card.
Conclusion:
- So far, I am reaching my goals of using artificial intelligence to detect objects.
- Working with Machine Learning has its easy and difficult points because there is no formula, but you build your own model and it must have an acceptable prediction rate.
- Now that this model works, the next step is to build the water sprayer system.