Skip to main content

9. 20-Class Object Detection Experiment

wmnologoAbout 2 min

Routine Code

import sensor, image, time, lcd
from maix import KPU
import gc

lcd.init()                          # Initialize LCD display
sensor.reset()                      # Reset and initialize camera
sensor.set_pixformat(sensor.RGB565) # Set camera output format to RGB565
sensor.set_framesize(sensor.QVGA)   # Set camera output size to QVGA (320x240)
sensor.skip_frames(time = 1000)     # Wait for camera to stabilize
clock = time.clock()                # Create a clock object for FPS calculation

# Detection model requires 320x256 input; initialize an image here
od_img = image.Image(size=(320,256))

obj_name = ("aeroplane","bicycle", "bird","boat","bottle","bus","car","cat","chair","cow","diningtable", "dog","horse", "motorbike","person","pottedplant", "sheep","sofa", "train", "tvmonitor")
anchor = (1.3221, 1.73145, 3.19275, 4.00944, 5.05587, 8.09892, 9.47112, 4.84053, 11.2364, 10.0071)
# Create a KPU object
kpu = KPU()
print("ready load model")
# Load model
#kpu.load_kmodel(0x300000, 1536936)
kpu.load_kmodel("/sd/KPU/voc20_object_detect/voc20_detect.kmodel")
# Initialize YOLO2
kpu.init_yolo2(anchor, anchor_num=5, img_w=320, img_h=240, net_w=320 , net_h=256 ,layer_w=10 ,layer_h=8, threshold=0.5, nms_value=0.2, classes=20)

i = 0
while True:
    i += 1
    print("cnt :", i)
    clock.tick()                        # Update FPS clock
    img = sensor.snapshot()             # Capture an image
    a = od_img.draw_image(img, 0,0)     # Write img to od_img at coordinates (0,0)
    od_img.pix_to_ai()                  # Convert RGB565 image to r8g8b8 format for AI inference
    kpu.run_with_output(od_img)         # Run KPU inference on input image
    dect = kpu.regionlayer_yolo2()      # YOLO2 post-processing
    fps = clock.fps()                   # Get FPS

    # Draw boxes and display object class
    if len(dect) > 0:
        print("dect:",dect)
        for l in dect :
            a = img.draw_rectangle(l[0],l[1],l[2],l[3], color=(0, 255, 0))
            a = img.draw_string(l[0],l[1], obj_name[l[4]], color=(0, 255, 0), scale=1.5)

    a = img.draw_string(0, 0, "%2.1ffps" %(fps), color=(0, 60, 128), scale=1.0)
    lcd.display(img)
    gc.collect()

# Deinitialize KPU object and release model memory
kpu.deinit()

Experiment Preparation

    1. First transfer model files to the TF card, then insert the TF card into the K210 module's TF card slot. For detailed steps, see Transfer Model Files to TF Card.
    1. Recognizable objects include: 0. aeroplane, 1. bicycle, 2. bird, 3. boat, 4. bottle, 5. bus, 6. car, 7. cat, 8. chair, 9. cow, 10. dining table, 11. dog, 12. horse, 13. motorbike, 14. person, 15. potted plant, 16. sheep, 17. sofa, 18. train, 19. TV monitor. Prepare any one of these objects.
    1. Connect the K210 to your computer via USB.
    1. Open CanMV IDE and run the routine code above.

Experiment Results

    1. When the K210 detects a trained object from the model, it draws a green rectangle around the object, displays the object name, and shows the current frame rate in the upper-left corner.
    1. Object detection results are printed in the IDE Serial Terminal.

Routine Code Explanation

Omitted. See comments in the routine code.

Copyright © 2026 wmnologo