YOLO Vision Code
YOLO 无界面人体检测：检测到人后保存图片
适合无人机或机载计算机无显示器运行，检测到人体后自动保存带框图片。
这些脚本用于 DRF450 或机载计算机上的 OpenCV + YOLOv8 视觉识别测试。建议先在桌面或树莓派本地验证摄像头索引、模型路径和性能，再把检测结果接入 MAVLink、自主飞行或任务触发逻辑。
脚本目标
后台读取摄像头并运行 YOLOv8，检测到 person 后按冷却时间保存标注图片到 detections 目录。
运行依赖： Python、OpenCV、Ultralytics YOLO。检测脚本默认模型路径为
/home/drobotics/yolov8n.pt，部署时请确认模型文件存在。学习重点
- 适合 SSH 后台运行和外场记录
- SAVE_COOLDOWN 防止连续保存过多图片
- 保存前绘制检测框，方便回看验证
关键参数
| 参数 | 当前设置 |
|---|---|
CAMERA_INDEX | 0 |
CONF_THRESHOLD | 0.5 |
SAVE_DIR | "detections" |
FRAME_SKIP | 2 # increase for more speed |
SAVE_COOLDOWN | 2 # seconds between saves (avoid spam) |
PERSON_CLASS_ID | 0 |
运行前检查
- 安装依赖：
pip install opencv-python ultralytics - 确认摄像头编号，必要时修改
CAMERA_INDEX或CAMERA_INDEXES。 - 确认 YOLO 模型路径，例如
/home/drobotics/yolov8n.pt。 - 在无人机上运行前，先单独验证摄像头、推理速度、保存路径和散热。
完整代码：yolo_detect_human.py
import cv2
import os
import time
from datetime import datetime
from ultralytics import YOLO
# =========================
# SETTINGS
# =========================
CAMERA_INDEX = 0
CONF_THRESHOLD = 0.5
SAVE_DIR = "detections"
FRAME_SKIP = 2 # increase for more speed
SAVE_COOLDOWN = 2 # seconds between saves (avoid spam)
# =========================
# INIT
# =========================
os.makedirs(SAVE_DIR, exist_ok=True)
print(" Loading YOLOv8...")
model = YOLO("/home/drobotics/yolov8n.pt")
PERSON_CLASS_ID = 0
cap = cv2.VideoCapture(CAMERA_INDEX)
if not cap.isOpened():
print(" Camera not opened")
exit()
print(" Running headless detection... saving only")
counter = 0
last_save_time = 0
# =========================
# LOOP
# =========================
while True:
ret, frame = cap.read()
if not ret:
print(" Frame error")
break
counter += 1
if counter % FRAME_SKIP != 0:
continue
# Resize for speed
frame = cv2.resize(frame, (640, 360))
results = model(frame, imgsz=320, verbose=False)[0]
person_detected = False
boxes_to_save = []
# =========================
# CHECK DETECTIONS
# =========================
for box in results.boxes:
cls_id = int(box.cls[0])
conf = float(box.conf[0])
if cls_id == PERSON_CLASS_ID and conf > CONF_THRESHOLD:
person_detected = True
x1, y1, x2, y2 = map(int, box.xyxy[0])
boxes_to_save.append((x1, y1, x2, y2, conf))
# =========================
# SAVE IMAGE IF PERSON FOUND
# =========================
current_time = time.time()
if person_detected and (current_time - last_save_time > SAVE_COOLDOWN):
timestamp = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
filename = os.path.join(SAVE_DIR, f"person_{timestamp}.jpg")
# draw boxes before saving
for (x1, y1, x2, y2, conf) in boxes_to_save:
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
cv2.putText(frame, f"Person {conf:.2f}",
(x1, y1 - 10),
cv2.FONT_HERSHEY_SIMPLEX,
0.5, (0, 255, 0), 2)
cv2.imwrite(filename, frame)
print(f" Saved detection: {filename}")
last_save_time = current_time
# =========================
# CLEANUP
# =========================
cap.release()
print(" Stopped")
下一步
先完成摄像头采集，再运行实时检测或无界面检测。后续可以把人体检测结果接入 MAVLink 任务逻辑，实现识别触发、悬停、返航、录像或地面站告警。