yolov8 物品分类案例
1 设备说明
myCobot 280 Arduino + Jetson Xavier NX开发板 + myCobot 垂直吸泵2.0 + myCobot 摄像头模组
2 硬件安装
3 原理说明
该程序结合了：
YOLOv8 目标识别模型（分类物体如“apple”、“banana”）
MyCobot 机械臂（带吸泵）
摄像头图像采集
GPIO 控制吸泵电磁阀
实现流程如下：
摄像头采集图像 → YOLOv8识别物体 → 获取物体中心点像素坐标 → 将像素坐标转换为真实机械臂坐标 → 控制机械臂移动并吸取 → 移动至分类投放点A/B/C/D → 释放物体 → 回初始位置
吸泵的启停通过 Jetson GPIO 控制，控制引脚为 BOARD 16。
4 程序运行
YOLOv8模型文件下载： best.pt
import cv2
import time
import threading
import traceback
from ultralytics import YOLO
from pymycobot import MyCobot280
import Jetson.GPIO as GPIO
GPIO.setmode(GPIO.BOARD)
GPIO.setup(16, GPIO.OUT)
class ElephantDetection:
def __init__(self, model_path):
self.model = YOLO(model_path) # 加载 YOLOv8 模型
def infer(self, frame):
results = self.model(frame)[0] # 对图像进行推理
result_image = frame.copy()
rect_list = []
classnames = []
for box in results.boxes:
x1, y1, x2, y2 = map(int, box.xyxy[0]) # 获取边框坐标
w, h = x2 - x1, y2 - y1
label = int(box.cls[0]) # 获取类别索引
conf = float(box.conf[0]) # 获取置信度
name = self.model.names[label] # 获取类别名称
# 添加矩形框和类别标签到图像上
cv2.rectangle(result_image, (x1, y1), (x2, y2), (0, 255, 0), 2)
cv2.putText(result_image, f'{name} {conf:.2f}', (x1, y1 - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
rect_list.append(((x1, y1), w, h, name, conf))
classnames.append(name)
return result_image, rect_list, classnames
class ElephantMotionControl:
def __init__(self):
self.task_completed = True
self.mc = MyCobot280('/dev/ttyTHS0', 1000000) # 初始化机械臂
time.sleep(1.5)
self.cam_coords = [197.8, -47.2, 236.2, -174.25, -0.66, -49.3] # 相机视角下的中心点坐标
self.init_pos = [4.83, -12.21, -60.9, -10.98, 3.51, -39.81]
self.pos_A = [62.4, -12.65, -64.95, -15.82, -1.14, -10.81] # 苹果投放点
self.pos_B = [39.55, -49.3, -21.79, -15.29, 5.36, -10.81] # 橙子投放点
self.pos_C = [-25.13, -39.28, -27.42, -33.75, -2.19, -9.05] # 香蕉投放点
self.pos_D = [-16.61, -49.65, -27.42, -5.09, 1.31, -9.05] # 葡萄投放点
self.speed = 50
self.coords_speed = 80
self.x_offset = 66
self.y_offset = -65
self.last_task_done_time = time.time()
if self.mc.get_fresh_mode() == 1:
self.mc.set_fresh_mode(0)
self.mc.send_angles(self.init_pos, self.speed) # 回到初始位置
self.check_position(self.init_pos, 0)
def check_position(self, data, ids, max_same_data_count=50):
"""阻塞等待直到机械臂到达指定位置（防止提前执行下一步）"""
try:
same_data_count = 0
last_data = None
while True:
res = self.mc.is_in_position(data, ids)
if data == last_data:
same_data_count += 1
else:
same_data_count = 0
last_data = data
if res == 1 or same_data_count >= max_same_data_count:
break
time.sleep(0.1)
except Exception:
print(traceback.format_exc())
# 开启吸泵
def pump_on(self):
GPIO.output(16, 0) # 开启吸泵（低电平有效）
time.sleep(0.05)
# 停止吸泵
def pump_off(self):
GPIO.output(16, 1) # 关闭吸泵
time.sleep(0.05)
def convert_to_real_coordinates(self, x_pixel, y_pixel, target_id):
if not self.task_completed:
return
self.task_completed = False
# 将像素坐标换算为实际坐标偏移量（单位：mm）
x = round(y_pixel * (60 / 170), 2)
y = round(-x_pixel * (60 / 170), 2)
# 更新目标坐标（机械臂将移动到这里进行抓取）
self.cam_coords[0] = 198.4 + x - self.x_offset
self.cam_coords[1] = -46.3 - y + self.y_offset
self.cam_coords[2] = 170 # Z 高度初始为安全高度
print(f"移动到新的坐标: {self.cam_coords} id--------->: {target_id}")
def move_robot():
try:
self.mc.send_coords(self.cam_coords, self.coords_speed, 1) # 降到物体上方
self.check_position(self.cam_coords, 1, 20)
self.cam_coords[2] = 120
self.mc.send_coords(self.cam_coords, self.coords_speed, 1) # 降下来
self.check_position(self.cam_coords, 1, 20)
self.pump_on() # 吸取物体
self.cam_coords[2] = 230
self.mc.send_coords(self.cam_coords, self.coords_speed, 1) # 抬起物体
self.check_position(self.cam_coords, 1, 20)
# 移动到目标位置（A/B/C/D）
if target_id == 1:
self.mc.send_angles(self.pos_A, self.speed)
self.check_position(self.pos_A, 0)
elif target_id == 2:
self.mc.send_angles(self.pos_B, self.speed)
self.check_position(self.pos_B, 0)
elif target_id == 3:
self.mc.send_angles(self.pos_C, self.speed)
self.check_position(self.pos_C, 0)
elif target_id == 4:
self.mc.send_angles(self.pos_D, self.speed)
self.check_position(self.pos_D, 0)
self.pump_off() # 释放
self.mc.send_angles(self.init_pos, self.speed) # 回到初始位
self.check_position(self.init_pos, 0)
self.last_task_done_time = time.time()
self.task_completed = True
self.cam_coords = None
while self.cam_coords is None:
self.cam_coords = self.mc.get_coords()
except Exception:
print(traceback.format_exc())
threading.Thread(target=move_robot).start()
def main():
detector = ElephantDetection("best.pt")
motion_control = ElephantMotionControl()
cap = cv2.VideoCapture(1, cv2.CAP_DSHOW)
last_result_image = None
while True:
ret, frame = cap.read()
if not ret:
continue
if motion_control.task_completed and time.time() - motion_control.last_task_done_time > 2.0:
result_img, rects, classnames = detector.infer(frame)
print(f"检测到 {len(rects)} 个目标，类别：{classnames}")
for rect in rects:
(x1, y1), w, h, label, conf = rect
center_x = x1 + w // 2
center_y = y1 + h // 2
print(f"{label} @ ({center_x}, {center_y})")
# 物品分类：apple/orange/banana1/grape → 对应投放点
if label in ["apple"]:
target_id = 1
elif label in ["orange"]:
target_id = 2
elif label in ["banana1"]:
target_id = 3
elif label in ["grape"]:
target_id = 4
else:
continue
# 发送坐标给机械臂控制模块进行移动
motion_control.convert_to_real_coordinates(center_x, center_y, target_id)
break # 每次只处理一个目标
last_result_image = result_img
cv2.imshow("Result", result_img)
else:
if last_result_image is not None:
cv2.imshow("Result", last_result_image)
else:
cv2.imshow("Result", frame)
if cv2.waitKey(1) & 0xFF == ord("q"):
break
cap.release()
cv2.destroyAllWindows()
if __name__ == "__main__":
main()