关于K210de那些事

1.K210如何在一个程序中实现不同的识别

1.1.怎么实现不同的识别模块在一个程序

        要实现不同的识别在同一个程序，而不重新加载模型，导致的内存溢出。是这一部分的关键，需要有外部改变条件，怎么改变条件呢，下面的串口便是（标题2）,话不多说直接上代码:

import sensor, image, time, lcd
from machine import UART
from fpioa_manager import fm
from maix import KPU
import gc


fm.register(7, fm.fpioa.UART1_TX, force=True) #10为tx接rx
fm.register(6, fm.fpioa.UART1_RX, force=True) #11为rx接tx
uart_A = UART(UART.UART1, 921600, 8, 1, 0, timeout=1000, read_buf_len=4096)


lcd.init()
lcd.clear(lcd.RED)

sensor.reset()
sensor.set_hmirror(0)
sensor.set_pixformat(sensor.RGB565)
sensor.set_framesize(sensor.QVGA)
sensor.skip_frames(time = 2000)

clock = time.clock()
def program_a():
    gc.collect()
    print('Running Program A.')
    od_img1 = image.Image(size=(320,256)) # 设置图像尺寸//-16.9

    anchor = (0.893, 1.463, 0.245, 0.389, 1.55, 2.58, 0.375, 0.594, 3.099, 5.038, 0.057, 0.090, 0.567, 0.904, 0.101, 0.160, 0.159, 0.255)
    kpu1 = KPU()
    kpu1.load_kmodel('/sd/yolo_face_detect.kmodel')  # 模型保存在SD卡中，从SD卡中直接加载模型

    kpu1.init_yolo2(anchor, anchor_num=9, img_w=320, img_h=240, net_w=320 , net_h=256 ,layer_w=10 ,layer_h=8, threshold=0.7, nms_value=0.3, classes=1)

    while True:
        clock.tick()
        img1 = sensor.snapshot()

        a = od_img1.draw_image(img1, 0,0)
        od_img1.pix_to_ai()
        # 对摄像头采集到的图像传递到KPU单元进行yolo模型运算//-25
        kpu1.run_with_output(od_img1)
        dect = kpu1.regionlayer_yolo2()
        fps = clock.fps()
        if len(dect) > 0:      # 识别到人脸
            print("dect:",dect)
            for l in dect :    # 画矩形框，框起人脸
                a = img1.draw_rectangle(l[0],l[1],l[2],l[3], color=(0, 255, 0))
        a = img1.draw_string(0, 0, "%2.1ffps" %(fps), color=(0, 60, 128), scale=2.0) #在LCD上显示帧率
        lcd.display(img1)
        compressed_img = img1.compress(quality=60)

        image_bytes = compressed_img.to_bytes()
        #print("send len: 压缩后的字节数 ", len(image_bytes))


        uart_A.write(image_bytes)
        #uart_B.write(image_bytes)
        gc.collect()      # 内存回收机制
        #kpu1.deinit()  # 释放KPU资源
        #del od_img1    # 删除图像对象
        #gc.collect()  # 再次进行内存回收//+16

        if uart_A.any():
            data = uart_A.read(128)
            print("Received data:", data.decode())
            if '0' in data.decode():
                print("EXIT A PROCESS")
                kpu1.deinit()  # 释放KPU资源
                del od_img1    # 删除图像对象
                gc.collect()  # 再次进行内存回收//+16
                break
            else:
                print("Invalid input from UART. Please send '1-5' for Program A-E")


def program_b():
    gc.collect()
    print('Running Program B.')
    #人脸检测模型需要320*256图输入，这里初始化一个image
    od_img2 = image.Image(size=(320,256), copy_to_fb=False)
    anchor = (0.156250, 0.222548, 0.361328, 0.489583, 0.781250, 0.983133, 1.621094, 1.964286, 3.574219, 3.94000)
    # 创建一个kpu对象，用于人脸检测
    kpu2= KPU()
    print("ready load model")
    # 加载模型
    kpu2.load_kmodel("/sd/detect_5.kmodel")
    # yolo2初始化
    kpu2.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=0.7, nms_value=0.4, classes=2)
    while True:
        #print("mem free:",gc.mem_free())
        clock.tick()                        # 更新计算帧率的clock
        img2 = sensor.snapshot()             # 拍照，获取一张图像
        a = od_img2.draw_image(img2, 0,0)     # 将img图像写到od_img图像的坐标（0,0）位置处
        od_img2.pix_to_ai()                  # 对rgb565的image生成ai运算需要的r8g8b8格式存储
        kpu2.run_with_output(od_img2)         # 对输入图像进行kpu运算
        dect = kpu2.regionlayer_yolo2()      # yolo2后处理
        fps = clock.fps()                   # 获取帧率
        # 画出人脸框，并判断是否带有口罩
        if len(dect) > 0:
            #uart_A.write('34')
            #print("dect:",dect)
            for l in dect :
                if l[4] :
                    a = img2.draw_rectangle(l[0],l[1],l[2],l[3], color=(0, 255, 0))
                    a = img2.draw_string(l[0],l[1]-24, "with mask", color=(0, 255, 0), scale=2)
                else:
                    a = img2.draw_rectangle(l[0],l[1],l[2],l[3], color=(255, 0, 0))
                    a = img2.draw_string(l[0],l[1]-24, "without mask", color=(255, 0, 0), scale=2)
        a = img2.draw_string(0, 0, "%2.1ffps" %(fps), color=(0, 60, 128), scale=2.0)
        lcd.display(img2)
        compressed_img2 = img2.compress(quality=40)
        image_bytes2 = compressed_img2.to_bytes()
        uart_A.write(image_bytes2)
        gc.collect()


        if uart_A.any():
            data = uart_A.read(128)
            print("Received data:", data.decode())
            if '0' in data.decode():
                print("EXIT B PROCESS")

                kpu2.deinit()  # 释放KPU资源
                del od_img2    # 删除图像对象
                gc.collect()  # 再次进行内存回收

                break
            else:
                print("Invalid input from UART. Please send '1-5' for Program A-E")

while True:
    gc.collect()      # 内存回收机制
    clock.tick()
    img = sensor.snapshot()
    lcd.display(img)
    print("71 mem free:",gc.mem_free())  # 查询剩余内存


    if uart_A.any():
        data = uart_A.read(128)
        print("Received data:", data.decode())
        if '1' in data.decode():

            program_a()
        elif '2' in data.decode():

            program_b()
        elif '3' in data.decode():

            program_c()
        elif '4' in data.decode():

            program_d()
        elif '5' in data.decode():

            program_e()
        else:
            print("Invalid input from UART. Please send '1-5' for Program A-E")

        上面的代码恒容易看出来，只需要通过输入端改变串口输入内容，同时保证占用内存资源的变量及时释放内存。

        下面将串口程序单独罗列便于查看。

2.K210串口使用进行数据上传

        直接上代码

from machine import UART
from fpioa_manager import fm

fm.register(7, fm.fpioa.UART1_TX, force=True) #10为tx接rx
fm.register(6, fm.fpioa.UART1_RX, force=True) #11为rx接tx
uart_A = UART(UART.UART1, 921600, 8, 1, 0, timeout=1000, read_buf_len=4096)

uart_A.write("hello")

加载包，初始化，使用，简单明了。