训练的对象检测模型的程序 如何在程序中使用传统的openmv的库来进一步处理图像

Viewed 29
/electronic_games training k230 ai

重现步骤
因为想要使用传统库处理 所以对摄像头初始化但这里的问题 在原先程序中他已经初始化过了 不过他初始化的获取的图像对象没有我需要方法 直接再加上一个初始化会报错显示已经初始化过了

期待结果和实际结果
<!- 你期待的结果是什么？实际看到的结果是什么？ -->预期可以正常初始化 实际报错

庐山派K230
软硬件版本信息

错误日志

尝试解决过程

补充材料

image.png

qi_liu1
1 Answers

参考下面的代码：

import os,sys,gc,time,ujson
from media.sensor import *
from media.display import *
from media.media import *
from libs.PlatTasks import DetectionApp
from libs.Utils import *
import nncase_runtime as nn
import ulab.numpy as np
import image

display_size=[800,480]
rgb565_size=[640,480]
rgb888p_size=[640,360]

sensor = Sensor()
sensor.reset()
#sensor.set_hmirror(False)
#sensor.set_vflip(False)
    
Display.init(Display.ST7701, width=display_size[0], height=display_size[1], osd_num=1, to_ide=True)
    
# 通道0直接给到显示VO，格式为YUV420
sensor.set_framesize(width = display_size[0], height = display_size[1],chn=CAM_CHN_ID_0)
sensor.set_pixformat(Sensor.YUV420SP,chn=CAM_CHN_ID_0)
# 通道1 给到openmv
sensor.set_framesize(width = rgb565_size[0], height = rgb565_size[1],chn=CAM_CHN_ID_1)
sensor.set_pixformat(Sensor.RGB565,chn=CAM_CHN_ID_1)
# 通道2给到AI做算法处理，格式为RGB888
sensor.set_framesize(width = rgb888p_size[0], height = rgb888p_size[1], chn=CAM_CHN_ID_2)
sensor.set_pixformat(Sensor.RGBP888, chn=CAM_CHN_ID_2)

# OSD图像初始化
osd_img = image.Image(display_size[0], display_size[1], image.ARGB8888)

sensor_bind_info = sensor.bind_info(x = 0, y = 0, chn = CAM_CHN_ID_0)
Display.bind_layer(**sensor_bind_info, layer = Display.LAYER_VIDEO1)

# media初始化
MediaManager.init()
# 启动sensor
sensor.run()


# Set root directory path for model and config
root_path = "/sdcard/mp_deployment_source/"

# Load deployment configuration
deploy_conf = read_json(root_path + "/deploy_config.json")
kmodel_path = root_path + deploy_conf["kmodel_path"]              # KModel path
labels = deploy_conf["categories"]                                # Label list
confidence_threshold = deploy_conf["confidence_threshold"]        # Confidence threshold
nms_threshold = deploy_conf["nms_threshold"]                      # NMS threshold
model_input_size = deploy_conf["img_size"]                        # Model input size
nms_option = deploy_conf["nms_option"]                            # NMS strategy
model_type = deploy_conf["model_type"]                            # Detection model type
anchors = []
if model_type == "AnchorBaseDet":
    anchors = deploy_conf["anchors"][0] + deploy_conf["anchors"][1] + deploy_conf["anchors"][2]

# Inference configuration
inference_mode = "video"                                          # Inference mode: 'video'
debug_mode = 0                                                    # Debug mode flag

# Initialize object detection application
det_app = DetectionApp(inference_mode,kmodel_path,labels,model_input_size,anchors,model_type,confidence_threshold,nms_threshold,rgb888p_size,display_size,debug_mode=debug_mode)

# Configure preprocessing for the model
det_app.config_preprocess()

# Main loop: capture, run inference, display results
while True:
    with ScopedTiming("total", 1):
        # ----------------这里拿到一帧rgb565的图像可以给openmv使用-----------
        img_rgb565=sensor.snapshot(chn=CAM_CHN_ID_1) 
        
        # ---------------------------------------------------------------
        
        # ----------------这里拿到一帧rgb888p的图像给AI模型使用---------------
        
        img_rgb888p = sensor.snapshot(chn=CAM_CHN_ID_2)       # Capture current frame
        img_np=img_rgb888p.to_numpy_ref()
        res = det_app.run(img_np)                        # Run inference
        det_app.draw_result(osd_img, res)          # Draw detection results
        Display.show_image(osd_img, 0, 0, Display.LAYER_OSD0)
        # ---------------------------------------------------------------
        gc.collect()                                  # Run garbage collection

# Cleanup: These lines will only run if the loop is interrupted (e.g., by an IDE break or external interruption)
det_app.deinit()                                      # De-initialize detection app
sensor.stop()
# deinit lcd
Display.deinit()
time.sleep_ms(50)
# deinit media buffer
MediaManager.deinit()
Wy0013018