2025-09-01 14:14:18 +08:00
|
|
|
|
from ultralytics import YOLO
|
|
|
|
|
|
import cv2
|
|
|
|
|
|
import numpy as np
|
|
|
|
|
|
import os
|
|
|
|
|
|
|
|
|
|
|
|
# 1. 加载模型
|
2025-09-05 14:29:33 +08:00
|
|
|
|
model = YOLO(r'/home/hx/yolo/ultralytics_yolo11-main/runs/train/exp_obb2/weights/best.pt')
|
2025-09-01 14:14:18 +08:00
|
|
|
|
|
|
|
|
|
|
# 2. 读取图像
|
2025-09-05 14:29:33 +08:00
|
|
|
|
img_path = r"/home/hx/yolo/output_masks/2.jpg"
|
2025-09-01 14:14:18 +08:00
|
|
|
|
img = cv2.imread(img_path)
|
|
|
|
|
|
|
|
|
|
|
|
if img is None:
|
|
|
|
|
|
print(f"❌ 错误:无法读取图像!请检查路径:{img_path}")
|
|
|
|
|
|
exit(1)
|
|
|
|
|
|
|
2025-09-05 14:29:33 +08:00
|
|
|
|
# 3. 预测(OBB 模式)
|
2025-09-01 14:14:18 +08:00
|
|
|
|
results = model(
|
|
|
|
|
|
img,
|
|
|
|
|
|
save=False,
|
2025-09-05 14:29:33 +08:00
|
|
|
|
imgsz=640,
|
|
|
|
|
|
conf=0.15,
|
|
|
|
|
|
mode='obb'
|
2025-09-01 14:14:18 +08:00
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
# 4. 获取结果并绘制
|
|
|
|
|
|
result = results[0]
|
2025-09-05 14:29:33 +08:00
|
|
|
|
annotated_img = result.plot()
|
2025-09-01 14:14:18 +08:00
|
|
|
|
|
2025-09-05 14:29:33 +08:00
|
|
|
|
# 5. 保存结果
|
2025-09-01 14:14:18 +08:00
|
|
|
|
output_dir = "./inference_results"
|
|
|
|
|
|
os.makedirs(output_dir, exist_ok=True)
|
|
|
|
|
|
filename = os.path.basename(img_path)
|
|
|
|
|
|
save_path = os.path.join(output_dir, "detected_" + filename)
|
|
|
|
|
|
cv2.imwrite(save_path, annotated_img)
|
|
|
|
|
|
print(f"✅ 推理结果已保存至: {save_path}")
|
|
|
|
|
|
|
2025-09-05 14:29:33 +08:00
|
|
|
|
# 6. 提取旋转框并计算 **两个框之间的夹角**
|
2025-09-01 14:14:18 +08:00
|
|
|
|
boxes = result.obb
|
|
|
|
|
|
if boxes is None or len(boxes) == 0:
|
|
|
|
|
|
print("❌ No objects detected.")
|
|
|
|
|
|
else:
|
|
|
|
|
|
print(f"✅ Detected {len(boxes)} object(s):")
|
2025-09-05 14:29:33 +08:00
|
|
|
|
directions = [] # 存储每个框的主方向(弧度),归一化到 [0, π)
|
2025-09-01 14:14:18 +08:00
|
|
|
|
|
|
|
|
|
|
for i, box in enumerate(boxes):
|
|
|
|
|
|
cls = int(box.cls.cpu().numpy()[0])
|
|
|
|
|
|
conf = box.conf.cpu().numpy()[0]
|
2025-09-05 14:29:33 +08:00
|
|
|
|
xywhr = box.xywhr.cpu().numpy()[0] # [cx, cy, w, h, r]
|
|
|
|
|
|
cx, cy, w, h, r_rad = xywhr
|
2025-09-01 14:14:18 +08:00
|
|
|
|
|
2025-09-05 14:29:33 +08:00
|
|
|
|
# 确定主方向(长边方向)
|
|
|
|
|
|
if w >= h:
|
|
|
|
|
|
direction = r_rad # 长边方向就是 r
|
|
|
|
|
|
else:
|
|
|
|
|
|
direction = r_rad + np.pi / 2 # 长边方向是 r + 90°
|
2025-09-01 14:14:18 +08:00
|
|
|
|
|
2025-09-05 14:29:33 +08:00
|
|
|
|
# 归一化到 [0, π)
|
|
|
|
|
|
direction = direction % np.pi
|
2025-09-01 14:14:18 +08:00
|
|
|
|
|
2025-09-05 14:29:33 +08:00
|
|
|
|
directions.append(direction)
|
|
|
|
|
|
angle_deg = np.degrees(direction)
|
|
|
|
|
|
print(f" Box {i+1}: Class: {cls}, Confidence: {conf:.3f}, 主方向: {angle_deg:.2f}°")
|
|
|
|
|
|
|
|
|
|
|
|
# ✅ 计算任意两个框之间的夹角(最小夹角,0° ~ 90°)
|
|
|
|
|
|
if len(directions) >= 2:
|
|
|
|
|
|
print("\n🔍 计算两个旋转框之间的夹角(主方向夹角):")
|
|
|
|
|
|
for i in range(len(directions)):
|
|
|
|
|
|
for j in range(i + 1, len(directions)):
|
|
|
|
|
|
dir1 = directions[i]
|
|
|
|
|
|
dir2 = directions[j]
|
|
|
|
|
|
|
|
|
|
|
|
# 计算方向差(取最小夹角,考虑周期性)
|
|
|
|
|
|
diff = abs(dir1 - dir2)
|
|
|
|
|
|
diff = min(diff, np.pi - diff) # 最小夹角(0 ~ π/2)
|
|
|
|
|
|
diff_deg = np.degrees(diff)
|
|
|
|
|
|
|
|
|
|
|
|
print(f" Box {i+1} 与 Box {j+1} 之间的夹角: {diff_deg:.2f}°")
|
2025-09-01 14:14:18 +08:00
|
|
|
|
else:
|
|
|
|
|
|
print("⚠️ 检测到少于两个目标,无法计算夹角。")
|
|
|
|
|
|
|
2025-09-05 14:29:33 +08:00
|
|
|
|
# 7. 显示图像
|
2025-09-01 14:14:18 +08:00
|
|
|
|
cv2.imshow("YOLO OBB Prediction", annotated_img)
|
|
|
|
|
|
cv2.waitKey(0)
|
|
|
|
|
|
cv2.destroyAllWindows()
|