【FireBeetle 2 ESP32-C5】API 连接与本地图像识别- Makelog(造物记)

【FireBeetle 2 ESP32-C5】API 连接与本地图像识别

本文介绍了 FireBeetle 2 ESP32-C5 开发套件基于网络 HTTP 协议 API 实现本地图像识别的项目设计。

项目介绍

准备工作：包括开发板固件烧录、模型下载、Python 库部署等；

流程图：包括开发板代码和网页服务器代码对应的流程图；

工程代码：包括开发板板端执行代码、网页服务器端代码等；

工程测试：包括程序运行、图片上传、识别结果查看等；

准备工作

下载 YOLOv8n.pt 模型，或程序自动从 GitHub 网站下载；

开发板烧录 MicroPython 固件；

电脑安装 Thonny IDE 软件，用来调试板端程序；

电脑安装 Python 软件，并部署所需库函数，如 opencv-python 等，便于模型识别与调用。

详见：【LattePanda Mu 开发套件】AI 视觉应用开发——物体识别- Makelog(造物记) .

流程图

网页服务器代码流程图

ESP32-C5 代码流程图

工程代码

包括 ESP32-C5 代码、网页服务器执行代码两部分。

ESP32-C5 代码

打开 Thonny IDE 软件，新建 api_object_recognition.py 工程文件，并添加如下代码

# ------ ESP32-C5 one-shot send local JPG to YOLO server ----------
import network
import urequests
import sys

# ---------- Wi-Fi ----------
SSID = "xxx"
PWD = "xxx"
SERVER = "http://192.168.31.101:5000/detect"

sta = network.WLAN(network.STA_IF)
sta.active(True)
sta.connect(SSID, PWD)
while not sta.isconnected():
   pass
print("Wi-Fi ok:", sta.ifconfig()[0])

# ---------- send target picture ----------
LOCAL_JPG = "desktop.jpg"

with open(LOCAL_JPG, "rb") as f:
   jpg = f.read()

boundary = "123456"
hdr = f"--{boundary}\r\n"
hdr += f'Content-Disposition: form-data; name="image"; filename="{LOCAL_JPG}"\r\n'
hdr += "Content-Type: image/jpeg\r\n\r\n"
end = f"\r\n--{boundary}--\r\n"
body = hdr.encode() + jpg + end.encode()

try:
   rsp = urequests.post(SERVER,
   data=body,
   headers={"Content-Type": f"multipart/form-data; boundary={boundary}"})
   data = rsp.json()
   print("objects:", data["objects"])
   print("result file:", data["result_file"])
except Exception as e:
   print("err:", e)

# ---------- finish ----------
print("done – program ends")
sys.exit()

保存代码；

网页服务器代码

在模型文件上级目录创建 yolo_http_server.py 文件，添加如下代码

"""
YOLOv8 HTTP inference server
POST /detect → multipart JPEG upload
GET /last → shows last annotated image in browser
"""
import os
import uuid
import time
from pathlib import Path
from flask import Flask, request, jsonify, send_file
import cv2
from ultralytics import YOLO

app = Flask(__name__)

# ---------- config -------------------------------------------------
MODEL_PATH = "models/yolov8n.pt" # will auto-download first run
STORAGE_DIR = Path("img") # same folder you used before
STORAGE_DIR.mkdir(exist_ok=True)
ALLOWED_EXT = {".jpg", ".jpeg", ".png"}

# ---------- warm-up model once -------------------------------------
print("Loading YOLOv8 model …")
model = YOLO(MODEL_PATH)
print("Model ready.")

# ---------- helpers ------------------------------------------------
def save_upload(multipart_file):
   """store original file and return Path object"""
   ext = Path(multipart_file.filename).suffix.lower()
   if ext not in ALLOWED_EXT:
   raise ValueError("Only .jpg .jpeg .png allowed")
   name = f"{int(time.time()*1000)}_{uuid.uuid4().hex[:8]}{ext}"
   path = STORAGE_DIR / name
   multipart_file.save(path)
   return path

def infer_and_annotate(image_path: Path):
   """run YOLOv8 + plot; return (results_list, annotated_cv2_image)"""
   img_bgr = cv2.imread(str(image_path))
   if img_bgr is None:
   raise RuntimeError("Cannot decode image")

   results = model(img_bgr) # ultralytics engine
   annotated_bgr = results[0].plot() # boxes + labels drawn

   # build serialisable list
   objects = []
   for box in results[0].boxes:
   cls = int(box.cls)
   conf = float(box.conf)
   xyxy = box.xyxy[0].tolist() # [x1,y1,x2,y2]
   objects.append({
   "label": model.names[cls],
   "conf": round(conf, 3),
   "box": [int(x) for x in xyxy]
   })
   return objects, annotated_bgr

# ---------- routes -------------------------------------------------
@app.route("/detect", methods=["POST"])
def detect():
   if "image" not in request.files:
   return jsonify(error="missing 'image' multipart field"), 400

   try:
   original_path = save_upload(request.files["image"])
   objects, annotated = infer_and_annotate(original_path)

   # save annotated image side-by-side
   annotated_name = original_path.stem + "_yolo.jpg"
   annotated_path = STORAGE_DIR / annotated_name
   cv2.imwrite(str(annotated_path), annotated)

   return jsonify(
   objects=objects,
   result_file=annotated_name,
   original_file=original_path.name
   )
   except Exception as e:
   return jsonify(error=str(e)), 500

@app.route("/last")
def last_image():
   """convenience: show most recent _yolo.jpg in browser"""
   files = sorted(STORAGE_DIR.glob("*_yolo.jpg"), key=os.path.getmtime, reverse=True)
   if not files:
   return "No annotated image yet", 404
   return send_file(files[0], mimetype="image/jpeg")

@app.route("/")
def root():
   return ("YOLOv8 inference server is running.
"
   "POST to /detect (multipart JPEG)
"
   "GET /last (newest annotated image)")
# ---------- entry --------------------------------------------------
if __name__ == "__main__":
   app.run(host="0.0.0.0", port=5000, threaded=True)

保存代码；

工程测试

命令行终端进入网页服务器代码所在文件夹；

终端执行 python yolo_http_server.py 指令；

打开终端提示的网址 http://192.168.31.109:5000/ 显示界面如下

表明模型加载成功且正在运行；

运行 Thonny IDE 并连接开发板，将目标图片 desktop.jpg 传输至板端根目录；

运行 api_object_recognition.py 程序，终端输出 WiFi 连接和物体识别结果；

打开网页 http://192.168.31.109:5000/last 查看图片识别结果；

总结

本文介绍了 FireBeetle 2 ESP32-C5 开发 HTTP 协议 API 实现本地图像识别的项目设计，包括网页服务器工程代码、板端执行代码、工程测试流程、识别结果查看等，为相关产品的开发设计和快速应用提供了参考。

【FireBeetle 2 ESP32-C5】API 连接与本地图像识别 简单