from libs.PipeLine import PipeLine, ScopedTiming
from libs.AIBase import AIBase
from libs.AI2D import Ai2d
import os
import ujson
from media.media import *
from time import *
import nncase_runtime as nn
import ulab.numpy as np
import time
import image
import aicube
import random
import gc
import sys
from libs.YbProtocol import YbProtocol
from ybUtils.YbUart import YbUart
# uart = None
uart = YbUart(baudrate=115200)
pto = YbProtocol()
# 自定义OCR检测类
# Custom OCR detection class
class OCRDetectionApp(AIBase):
def __init__(self, kmodel_path, model_input_size, mask_threshold=0.5, box_threshold=0.2, rgb888p_size=[224,224], display_size=[1920,1080], debug_mode=0):
"""
初始化OCR检测应用
Initialize the OCR detection application
参数:
kmodel_path: 模型路径 / Model path
model_input_size: 模型输入大小 / Model input size
mask_threshold: 掩码阈值，用于区分前景和背景 / Mask threshold for foreground/background separation
box_threshold: 边界框阈值，用于确定检测结果 / Box threshold for detection results
rgb888p_size: 输入图像大小 / Input image size
display_size: 显示大小 / Display size
debug_mode: 调试模式级别 / Debug mode level
"""
super().__init__(kmodel_path, model_input_size, rgb888p_size, debug_mode)
self.kmodel_path = kmodel_path
# 模型输入分辨率 / Model input resolution
self.model_input_size = model_input_size
# 分类阈值 / Classification thresholds
self.mask_threshold = mask_threshold # 掩码阈值 / Mask threshold
self.box_threshold = box_threshold # 框阈值 / Box threshold
# sensor给到AI的图像分辨率，宽度16字节对齐 / Image resolution from sensor to AI, width aligned to 16 bytes
self.rgb888p_size = [ALIGN_UP(rgb888p_size[0], 16), rgb888p_size[1]]
# 显示分辨率 / Display resolution
self.display_size = [ALIGN_UP(display_size[0], 16), display_size[1]]
self.debug_mode = debug_mode
# Ai2d实例，用于实现模型预处理 / Ai2d instance for model preprocessing
self.ai2d = Ai2d(debug_mode)
# 设置Ai2d的输入输出格式和类型 / Set input and output formats and types for Ai2d
self.ai2d.set_ai2d_dtype(nn.ai2d_format.NCHW_FMT, nn.ai2d_format.NCHW_FMT, np.uint8, np.uint8)
# 配置预处理操作，这里使用了pad和resize
# Configure preprocessing operations, using pad and resize here
# Ai2d支持crop/shift/pad/resize/affine，具体代码请打开/sdcard/app/libs/AI2D.py查看
# Ai2d supports crop/shift/pad/resize/affine, see /sdcard/app/libs/AI2D.py for details
def config_preprocess(self, input_image_size=None):
with ScopedTiming("set preprocess config", self.debug_mode > 0):
# 初始化ai2d预处理配置，默认为sensor给到AI的尺寸，您可以通过设置input_image_size自行修改输入尺寸
# Initialize ai2d preprocessing config, default is the size from sensor to AI
# You can modify the input size by setting input_image_size
ai2d_input_size = input_image_size if input_image_size else self.rgb888p_size
top, bottom, left, right = self.get_padding_param()
self.ai2d.pad([0, 0, 0, 0, top, bottom, left, right], 0, [0, 0, 0])
self.ai2d.resize(nn.interp_method.tf_bilinear, nn.interp_mode.half_pixel)
self.ai2d.build([1, 3, ai2d_input_size[1], ai2d_input_size[0]], [1, 3, self.model_input_size[1], self.model_input_size[0]])
# 自定义当前任务的后处理
# Custom post-processing for the current task
def postprocess(self, results):
with ScopedTiming("postprocess", self.debug_mode > 0):
# chw2hwc 将数据从[channel,height,width]转换为[height,width,channel]格式
# Convert data from [channel,height,width] to [height,width,channel] format
hwc_array = self.chw2hwc(self.cur_img)
# 这里使用了aicube封装的接口ocr_post_process做后处理
# Using aicube's ocr_post_process interface for post-processing
# 返回的det_boxes结构为[[crop_array_nhwc,[p1_x,p1_y,p2_x,p2_y,p3_x,p3_y,p4_x,p4_y]],...]
# The structure of returned det_boxes is [[crop_array_nhwc,[p1_x,p1_y,p2_x,p2_y,p3_x,p3_y,p4_x,p4_y]],...]
det_boxes = aicube.ocr_post_process(results[0][:,:,:,0].reshape(-1),
hwc_array.reshape(-1),
self.model_input_size,
self.rgb888p_size,
self.mask_threshold,
self.box_threshold)
return det_boxes
# 计算padding参数
# Calculate padding parameters
def get_padding_param(self):
# 右padding或下padding / Right padding or bottom padding
dst_w = self.model_input_size[0]
dst_h = self.model_input_size[1]
input_width = self.rgb888p_size[0]
input_high = self.rgb888p_size[1]
# 计算缩放比例 / Calculate scaling ratios
ratio_w = dst_w / input_width
ratio_h = dst_h / input_high
# 选择较小的缩放比例，保持原始宽高比 / Choose the smaller ratio to maintain the original aspect ratio
if ratio_w < ratio_h:
ratio = ratio_w
else:
ratio = ratio_h
# 计算缩放后的新尺寸 / Calculate new dimensions after scaling
new_w = (int)(ratio * input_width)
new_h = (int)(ratio * input_high)
# 计算padding量 / Calculate padding amounts
dw = (dst_w - new_w) / 2
dh = (dst_h - new_h) / 2
# 四边的padding值 / Padding values for all four sides
top = (int)(round(0))
bottom = (int)(round(dh * 2 + 0.1))
left = (int)(round(0))
right = (int)(round(dw * 2 - 0.1))
return top, bottom, left, right
# chw2hwc 转换通道顺序
# chw2hwc converts channel order
def chw2hwc(self, features):
# 从[channel,height,width]转换为[height,width,channel]
# Convert from [channel,height,width] to [height,width,channel]
ori_shape = (features.shape[0], features.shape[1], features.shape[2])
c_hw_ = features.reshape((ori_shape[0], ori_shape[1] * ori_shape[2]))
hw_c_ = c_hw_.transpose()
new_array = hw_c_.copy()
hwc_array = new_array.reshape((ori_shape[1], ori_shape[2], ori_shape[0]))
# 释放中间变量，降低内存占用 / Release intermediate variables to reduce memory usage
del c_hw_
del hw_c_
del new_array
return hwc_array
# 自定义OCR识别任务类
# Custom OCR recognition task class
class OCRRecognitionApp(AIBase):
def __init__(self, kmodel_path, model_input_size, dict_path, rgb888p_size=[1920,1080], display_size=[1920,1080], debug_mode=0):
"""
初始化OCR识别应用
Initialize OCR recognition application
参数:
kmodel_path: 模型路径 / Model path
model_input_size: 模型输入大小 / Model input size
dict_path: 字典文件路径，用于将识别结果映射为文字 / Dictionary file path for mapping recognition results to text
rgb888p_size: 输入图像大小 / Input image size
display_size: 显示大小 / Display size
debug_mode: 调试模式级别 / Debug mode level
"""
super().__init__(kmodel_path, model_input_size, rgb888p_size, debug_mode)
# kmodel路径 / kmodel path
self.kmodel_path = kmodel_path
# 识别模型输入分辨率 / Recognition model input resolution
self.model_input_size = model_input_size
self.dict_path = dict_path
# sensor给到AI的图像分辨率，宽16字节对齐 / Image resolution from sensor to AI, width aligned to 16 bytes
self.rgb888p_size = [ALIGN_UP(rgb888p_size[0], 16), rgb888p_size[1]]
# 视频输出VO分辨率，宽16字节对齐 / Video output VO resolution, width aligned to 16 bytes
self.display_size = [ALIGN_UP(display_size[0], 16), display_size[1]]
# debug模式 / debug mode
self.debug_mode = debug_mode
self.dict_word = None
# 读取OCR的字典 / Read OCR dictionary
self.read_dict()
self.ai2d = Ai2d(debug_mode)
self.ai2d.set_ai2d_dtype(nn.ai2d_format.RGB_packed, nn.ai2d_format.NCHW_FMT, np.uint8, np.uint8)
# 配置预处理操作
# Configure preprocessing operations
def config_preprocess(self, input_image_size=None, input_np=None):
with ScopedTiming("set preprocess config", self.debug_mode > 0):
ai2d_input_size = input_image_size if input_image_size else self.rgb888p_size
top, bottom, left, right = self.get_padding_param(ai2d_input_size, self.model_input_size)
self.ai2d.pad([0, 0, 0, 0, top, bottom, left, right], 0, [0, 0, 0])
self.ai2d.resize(nn.interp_method.tf_bilinear, nn.interp_mode.half_pixel)
# 如果传入input_np，输入shape为input_np的shape,如果不传入，输入shape为[1,3,ai2d_input_size[1],ai2d_input_size[0]]
# If input_np is provided, use its shape as input shape
# Otherwise, use [1,3,ai2d_input_size[1],ai2d_input_size[0]] as input shape
self.ai2d.build([input_np.shape[0], input_np.shape[1], input_np.shape[2], input_np.shape[3]],
[1, 3, self.model_input_size[1], self.model_input_size[0]])
# 自定义后处理，results是模型输出的array列表
# Custom post-processing, results is the array list output by the model
def postprocess(self, results):
with ScopedTiming("postprocess", self.debug_mode > 0):
# 获取每一列的最高概率对应的字符索引 / Get character index corresponding to highest probability for each column
preds = np.argmax(results[0], axis=2).reshape((-1))
output_txt = ""
for i in range(len(preds)):
# 当前识别字符不是字典的最后一个字符并且和前一个字符不重复（去重），加入识别结果字符串
# Add character to result string if it's not the last character in dictionary
# and not a duplicate of the previous character (deduplication)
if preds[i] != (len(self.dict_word) - 1) and (not (i > 0 and preds[i - 1] == preds[i])):
output_txt = output_txt + self.dict_word[preds[i]]
return output_txt
# 计算padding参数
# Calculate padding parameters
def get_padding_param(self, src_size, dst_size):
# 右padding或下padding / Right or bottom padding
dst_w = dst_size[0]
dst_h = dst_size[1]
input_width = src_size[0]
input_high = src_size[1]
# 计算缩放比例 / Calculate scaling ratios
ratio_w = dst_w / input_width
ratio_h = dst_h / input_high
# 选择较小的缩放比例，保持原始宽高比 / Choose smaller ratio to maintain aspect ratio
if ratio_w < ratio_h:
ratio = ratio_w
else:
ratio = ratio_h
# 计算缩放后的新尺寸 / Calculate new dimensions after scaling
new_w = (int)(ratio * input_width)
new_h = (int)(ratio * input_high)
# 计算padding量 / Calculate padding amounts
dw = (dst_w - new_w) / 2
dh = (dst_h - new_h) / 2
# 四边的padding值 / Padding values for all four sides
top = (int)(round(0))
bottom = (int)(round(dh * 2 + 0.1))
left = (int)(round(0))
right = (int)(round(dw * 2 - 0.1))
return top, bottom, left, right
# 读取字典文件
# Read dictionary file
def read_dict(self):
if self.dict_path != "":
with open(dict_path, 'r') as file:
line_one = file.read(100000)
line_list = line_one.split("\r\n")
# 创建字典，将索引映射到字符 / Create dictionary mapping indices to characters
self.dict_word = {num: char.replace("\r", "").replace("\n", "") for num, char in enumerate(line_list)}
# OCR检测和识别的组合类
# Combined class for OCR detection and recognition
class OCRDetRec:
def __init__(self, ocr_det_kmodel, ocr_rec_kmodel, det_input_size, rec_input_size, dict_path,
mask_threshold=0.25, box_threshold=0.3, rgb888p_size=[1920,1080], display_size=[1920,1080], debug_mode=0):
"""
初始化OCR检测和识别的组合应用
Initialize combined OCR detection and recognition application
参数:
ocr_det_kmodel: OCR检测模型路径 / OCR detection model path
ocr_rec_kmodel: OCR识别模型路径 / OCR recognition model path
det_input_size: 检测模型输入大小 / Detection model input size
rec_input_size: 识别模型输入大小 / Recognition model input size
dict_path: 字典文件路径 / Dictionary file path
mask_threshold: 掩码阈值 / Mask threshold
box_threshold: 框阈值 / Box threshold
rgb888p_size: 输入图像大小 / Input image size
display_size: 显示大小 / Display size
debug_mode: 调试模式级别 / Debug mode level
"""
# OCR检测模型路径 / OCR detection model path
self.ocr_det_kmodel = ocr_det_kmodel
# OCR识别模型路径 / OCR recognition model path
self.ocr_rec_kmodel = ocr_rec_kmodel
# OCR检测模型输入分辨率 / OCR detection model input resolution
self.det_input_size = det_input_size
# OCR识别模型输入分辨率 / OCR recognition model input resolution
self.rec_input_size = rec_input_size
# 字典路径 / Dictionary path
self.dict_path = dict_path
# 置信度阈值 / Confidence threshold
self.mask_threshold = mask_threshold
# nms阈值 / NMS threshold
self.box_threshold = box_threshold
# sensor给到AI的图像分辨率，宽16字节对齐 / Image resolution from sensor to AI, width aligned to 16 bytes
self.rgb888p_size = [ALIGN_UP(rgb888p_size[0], 16), rgb888p_size[1]]
# 视频输出VO分辨率，宽16字节对齐 / Video output resolution, width aligned to 16 bytes
self.display_size = [ALIGN_UP(display_size[0], 16), display_size[1]]
# debug_mode模式 / debug_mode mode
self.debug_mode = debug_mode
# 创建检测和识别实例 / Create detection and recognition instances
self.ocr_det = OCRDetectionApp(self.ocr_det_kmodel, model_input_size=self.det_input_size,
mask_threshold=self.mask_threshold, box_threshold=self.box_threshold,
rgb888p_size=self.rgb888p_size, display_size=self.display_size, debug_mode=0)
self.ocr_rec = OCRRecognitionApp(self.ocr_rec_kmodel, model_input_size=self.rec_input_size,
dict_path=self.dict_path, rgb888p_size=self.rgb888p_size,
display_size=self.display_size)
# 配置检测模型预处理 / Configure detection model preprocessing
self.ocr_det.config_preprocess()
# run函数，执行推理
# Run function for inference
def run(self, input_np):
# 先进行OCR检测 / First perform OCR detection
det_res = self.ocr_det.run(input_np)
boxes = []
ocr_res = []
for det in det_res:
# 对得到的每个检测框执行OCR识别 / Perform OCR recognition on each detected box
self.ocr_rec.config_preprocess(input_image_size=[det[0].shape[2], det[0].shape[1]], input_np=det[0])
ocr_str = self.ocr_rec.run(det[0])
ocr_res.append(ocr_str)
boxes.append(det[1])
# 执行垃圾回收，减少内存占用 / Perform garbage collection to reduce memory usage
gc.collect()
return boxes, ocr_res
# 绘制OCR检测识别效果
# Draw OCR detection and recognition results
def draw_result(self, pl, det_res, rec_res):
# 清除叠加层 / Clear overlay layer
pl.osd_img.clear()
if det_res:
# 循环绘制所有检测到的框 / Loop through all detected boxes
for j in range(len(det_res)):
# 将原图的坐标点转换成显示的坐标点，循环绘制四条直线，得到一个矩形框
# Convert coordinates from original image to display coordinates
# Draw four lines to form a rectangle
for i in range(4):
# 坐标转换 / Coordinate conversion
x1 = det_res[j][(i * 2)] / self.rgb888p_size[0] * self.display_size[0]
y1 = det_res[j][(i * 2 + 1)] / self.rgb888p_size[1] * self.display_size[1]
x2 = det_res[j][((i + 1) * 2) % 8] / self.rgb888p_size[0] * self.display_size[0]
y2 = det_res[j][((i + 1) * 2 + 1) % 8] / self.rgb888p_size[1] * self.display_size[1]
# 绘制线段 / Draw line segment
pl.osd_img.draw_line((int(x1), int(y1), int(x2), int(y2)), color=(255, 0, 0, 255), thickness=5)
# 在框上方绘制识别文本 / Draw recognized text above the box
pl.osd_img.draw_string_advanced(int(x1), int(y1), 32, rec_res[j], color=(0, 0, 255))
pto_data = pto.get_ocr_rec_data(rec_res[j])
uart.send(pto_data)
print(pto_data)
if __name__ == "__main__":
# 显示模式，默认"hdmi"，可以选择"hdmi"和"lcd"，k230d受限内存不支持
# Display mode, default is "hdmi", can choose between "hdmi" and "lcd", k230d with limited memory doesn't support
display_mode = "lcd"
if display_mode == "hdmi":
display_size = [1920, 1080]
else:
display_size = [640, 480]
# OCR检测模型路径 / OCR detection model path
ocr_det_kmodel_path = "/sdcard/kmodel/ocr_det_int16.kmodel"
# OCR识别模型路径 / OCR recognition model path
ocr_rec_kmodel_path = "/sdcard/kmodel/ocr_rec_int16.kmodel"
# 其他参数 / Other parameters
dict_path = "/sdcard/utils/dict.txt"
rgb888p_size = [640, 360]
ocr_det_input_size = [640, 640]
ocr_rec_input_size = [512, 32]
mask_threshold = 0.25
box_threshold = 0.3
# 初始化PipeLine，只关注传给AI的图像分辨率，显示的分辨率
# Initialize PipeLine, focusing only on image resolution for AI and display
pl = PipeLine(rgb888p_size=rgb888p_size, display_size=display_size, display_mode=display_mode)
pl.create()
# 创建OCR检测识别实例 / Create OCR detection and recognition instance
ocr = OCRDetRec(ocr_det_kmodel_path, ocr_rec_kmodel_path, det_input_size=ocr_det_input_size,
rec_input_size=ocr_rec_input_size, dict_path=dict_path, mask_threshold=mask_threshold,
box_threshold=box_threshold, rgb888p_size=rgb888p_size, display_size=display_size)
while True:
# 计时整个处理流程 / Time the entire processing flow
with ScopedTiming("total", 0):
img = pl.get_frame() # 获取当前帧 / Get current frame
det_res, rec_res = ocr.run(img) # 推理当前帧 / Inference on current frame
ocr.draw_result(pl, det_res, rec_res) # 绘制当前帧推理结果 / Draw inference results for current frame
pl.show_image() # 展示当前帧推理结果 / Display inference results for current frame
# 执行垃圾回收，减少内存占用 / Perform garbage collection to reduce memory usage
gc.collect()
# 释放资源 / Release resources
ocr.ocr_det.deinit()
ocr.ocr_rec.deinit()
pl.destroy()
让程序能准确识别数字
本文详细解析了如何在优快云上发布一篇高质量的技术博客，包括标题、标签和内容的优化策略，旨在帮助作者提高文章的可见性和吸引力。
扫一扫
1530
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
