yolov8obb&opencv

Question

使用此代码运行完全部视频后TAR文件夹中只存了五张图片，远少于实际目标数量

import time
import cv2
from ultralytics import YOLO
from ultralytics.utils.plotting import save_one_box as crop
import imutils
import numpy as np
import os
import shutil

model = YOLO('best.pt')

video_path = "D:/WIN_20230907_21_08_23_Pro.mp4"
cap = cv2.VideoCapture(video_path)
confi = 0.2

if not os.path.exists('TAR'):
    os.mkdir('TAR')

while cap.isOpened():
    success, frame = cap.read()

    confi = round(confi, 2)
    if confi <= 0.2:
        confi = 0.2
    elif confi >= 0.8:
        confi = 0.8

    if success:
        Results = model(frame, conf=confi, save=False, show=False)
        annotated_frame = Results[0].plot()
        cv2.imshow("YOLOv8 Tracking", annotated_frame)
        r = Results[0]
        ori_h, ori_w = r.orig_shape

        if Results:
            obbs = r.obb.xywhr.tolist()
            for i, obb in enumerate(obbs):
                xc, yc, w, h, angle = obb
                print(xc, yc, w, h, angle)

                roM = cv2.getRotationMatrix2D((xc, yc), angle, 1)
                ro_frame = cv2.warpAffine(frame, roM, (ori_w, ori_h))
                cropped_frame = cv2.getRectSubPix(ro_frame, (int(w), int(h)), (xc, yc))
                filename = f"TAR/cropped_{i}.jpg"
                cv2.imwrite(filename, cropped_frame)
        else:
            bkg = np.ones((256, 1280, 3), dtype=np.uint8)
            bkg *= 200
            annotated_frame = np.vstack((annotated_frame, bkg))

        if cv2.waitKey(1) & 0xFF == ord("q"):
            break
    else:
        # Break the loop if the end of the video is reached
        break

    # Release the video capture object and close the display window
cap.release()
cv2.destroyAllWindows()

Answer 1

下午好🌅🌅🌅
本答案参考ChatGPT-3.5

根据代码中注释的内容，该代码主要是使用YOLOv8物体检测模型对视频中的目标进行检测和跟踪，并保存特定目标的裁剪图像。但是，使用该代码运行完全部视频后，TAR文件夹中只存了五张图片，远少于实际目标数量。

解决方案：

1.检查代码中的阈值(confi)设置。可以尝试使用不同的阈值进行测试，以获取更多的目标图像文件。

2.检查代码中的模型加载是否成功。将model(frame, conf=confi, save=False, show=False)返回的Results对象进行打印并检查是否存在检测到的目标。

3.当前代码仅输出第一个标签的检测结果(Results [0])。可以尝试更改代码以输出更多标签的检测结果并检查是否有更多的裁剪图像文件。例如，可以在"for"循环中打开注释代码，同时保存其他标签的检测结果。

# res = Results.xyxy[Results.pred[:, -1] != 0]  # XYXY, cls label == 0 by default
# for *xyxy, conf, cls in reversed(res):
#     c = int(cls)  # integer class
#     label = None if c == 0 else model.names[c]  # apply a label
#     plot_xyxy(image, [xyxy], None, colors[c], label, line_thickness=hyp['thickness'])

4.更改代码中的保存文件名格式，以避免因重复命名而覆盖文件。例如，将文件名更改为在上传时间添加一个时间戳，可以避免出现同名文件。

5.检查代码中的图像分辨率是否正确。根据目标的绝对坐标中心(即 x,y 坐标)来裁剪图像时，需要将旋转之后的图像坐标系与原始帧对齐。如果未对图像和目标尺寸进行修正，可能会导致计算的目标位置偏离实际位置。

6.对于检测到的目标，可以通过使用cv2.imwrite函数将其保存在本地。或者可以将目标的原始帧保存下来，以保存完整的目标信息，而不仅仅是裁剪后的部分。

修改后的代码如下。修改后的代码将识别到的所有目标都保存在本地：

import time
import cv2
from ultralytics import YOLO
from ultralytics.utils.plotting import save_one_box as crop
import imutils
import numpy as np
import os
import shutil

model = YOLO('best.pt')

video_path = "D:/WIN_20230907_21_08_23_Pro.mp4"
cap = cv2.VideoCapture(video_path)
confi = 0.2

if not os.path.exists('TAR'):
    os.mkdir('TAR')
if not os.path.exists('all_objects'):
    os.mkdir('all_objects')

while cap.isOpened():
    success, frame = cap.read()

    confi = round(confi, 2)
    if confi <= 0.2:
        confi = 0.2
    elif confi >= 0.8:
        confi = 0.8

    if success:
        Results = model(frame, conf=confi, save=False, show=False)
        annotated_frame = Results[0].plot()
        cv2.imshow("YOLOv8 Tracking", annotated_frame)
        if Results:
            for i, r in enumerate(Results):
                obbs = r.obb.xywhr.tolist()
                ori_h, ori_w = r.orig_shape
                for obb in obbs:
                    xc, yc, w, h, angle = obb
                    roM = cv2.getRotationMatrix2D((xc, yc), angle, 1)
                    ro_frame = cv2.warpAffine(frame, roM, (ori_w, ori_h))
                    cropped_frame = cv2.getRectSubPix(ro_frame, (int(w), int(h)), (xc, yc))
                    timestamp = time.strftime("%Y%m%d-%H%M%S")
                    filename = f"TAR/image_{timestamp}_{i}.jpg"
                    cv2.imwrite(filename, cropped_frame)
                    all_filename = f"all_objects/image_{timestamp}_{i}_{h}_{w}_{xc}_{yc}.jpg"
                    cv2.imwrite(all_filename, frame)
        else:
            bkg = np.ones((256, 1280, 3), dtype=np.uint8)
            bkg *= 200
            annotated_frame = np.vstack((annotated_frame, bkg))

        if cv2.waitKey(1) & 0xFF == ord("q"):
            break
    else:
        # Break the loop if the end of the video is reached
        break

    # Release the video capture object and close the display window
cap.release()
cv2.destroyAllWindows()