yolov8 目标检测与跟踪

参考：
参考：
https://github.com/ultralytics/ultralytics
https://github.com/TommyZihao/Train_Custom_Dataset/blob/main/%E7%9B%AE%E6%A0%87%E8%BF%BD%E8%B8%AA/%E5%85%AC%E5%BC%80%E8%AF%BE/

https://www.rstk.cn/news/42041.html?action=onClick

*** 跟踪与检测都是用的YOLOv8目标检测一样的权重

1、命令行运行

视频下载：

人流量视频：
https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20220418-mmtracking/data/mot_people_short.mp4
车流量视频：
https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230502-YOLO/videos/bridge-short.mp4

可能出现报错OMP: Error #15: Initializing libiomp5md.dll, but found libiomp5md.dll already initialized.

解决方法参考：https://blog.csdn.net/qq_37164776/article/details/126832303
需要去env环境里把其中一个删除再运行

可能出现报错 im = torch.from_numpy(im)

RuntimeError: Numpy is not available

##升级numpy
从1.22.0到1.23.5解决pip install numpy==1.23.5  -i http://pypi.douban.com/simple --trusted-host pypi.douban.com --user

##检测
yolo detect predict model=C:\Users\lonng\Downloads\yolov8s.pt source=bus.jpg ##图片yolo detect predict model=C:\Users\lonng\Downloads\yolov8s.pt source=C:\Users\lonng\Desktop\rdkit_learn\opencv2\videos\bridge-short.mp4  show=True save=True   ##视频##调摄像头
yolo detect predict model=C:\Users\lonng\Downloads\yolov8s.pt source=0  show=True

注意：跟踪命令行没有predict，只有track

##跟踪，文件路径不需要加引号
yolo track model=C:\Users\lonng\Downloads\yolov8s.pt  source=C:\Users\lonng\Desktop\rdkit_learn\opencv2\videos\mot_people_short.mp4 show=True save=Trueyolo track  model=C:\Users\lonng\Downloads\yolov8s.pt source=0  show=True ##摄像头

运行解释会保存到runs目录下

2、python代码运行

跟踪完整代码：

import time
from tqdm import tqdmimport cv2
import numpy as npfrom ultralytics import YOLOimport supervision as svmodel = YOLO(r'C:\Users\lonng\Downloads\yolov8s.pt')VIDEO_PATH = 'videos/bridge-short.mp4'##指定统计区域
polygons = [np.array([[600, 680], [927, 680], [851, 950], [42, 950]]),np.array([[987, 680], [1350, 680], [1893, 950], [1015, 950]])
]zones = [sv.PolygonZone(polygon=polygon, frame_resolution_wh=video_info.resolution_wh) for polygon in polygons]
# 配色方案
colors = sv.ColorPalette.default()
# 区域可视化，每个区域配一个 PolygonZoneAnnotator
zone_annotators = [sv.PolygonZoneAnnotator(zone=zone, color=colors.by_idx(index), thickness=6, text_thickness=12, text_scale=4)for index, zone in enumerate(zones)
]
# 目标检测可视化，每个区域配一个 BoxAnnotator
box_annotators = [sv.BoxAnnotator(color=colors.by_idx(index), thickness=2, text_thickness=4, text_scale=2)for index in range(len(polygons))
]##每帧处理
def process_frame(frame: np.ndarray, i) -> np.ndarray:'''输入 bgr 的 numpy array输出 bgr 的 numpy array'''# YOLOV8 推理预测results = model(frame, imgsz=1280, verbose=False, show=False, device='cuda:0')[0]# 用 supervision 解析预测结果detections = sv.Detections.from_yolov8(results)# 遍历每个区域对应的所有 Annotatorfor zone, zone_annotator, box_annotator in zip(zones, zone_annotators, box_annotators):# 判断目标是否在区域内mask = zone.trigger(detections=detections)# 筛选出在区域内的目标detections_filtered = detections[mask]# 画框frame = box_annotator.annotate(scene=frame, detections=detections_filtered, skip_label=True)# 画区域，并写区域内目标个数frame = zone_annotator.annotate(scene=frame)# 更新进度条pbar.update(1)return frame## 视频每帧预测
filehead = VIDEO_PATH.split('/')[-1]
OUT_PATH = "out-" + fileheadwith tqdm(total=video_info.total_frames-1) as pbar:sv.process_video(source_path=VIDEO_PATH, target_path=OUT_PATH, callback=process_frame)

out-bridge-short

跟踪

参考：https://github.com/TommyZihao/Train_Custom_Dataset/blob/main/%E7%9B%AE%E6%A0%87%E8%BF%BD%E8%B8%AA/%E5%85%AC%E5%BC%80%E8%AF%BE/%E3%80%90D1%E3%80%91%E8%BD%A6%E6%B5%81%E8%B6%8A%E7%BA%BF%E8%AE%A1%E6%95%B0-%E8%A7%86%E9%A2%91%E9%A2%84%E6%B5%8B-%E7%AE%80%E5%8D%95%E5%9C%BA%E6%99%AF.ipynb

from ultralytics import YOLOimport supervision as svmodel = YOLO(r'C:\Users\lonng\Downloads\yolov8s.pt')###track格式 model.track
for result in model.track(source=input_path, show=False, stream=True, verbose=False, device='cuda:0'):frame = result.orig_img# 用 supervision 解析预测结果detections = sv.Detections.from_yolov8(result)## 过滤掉某些类别# detections = detections[(detections.class_id != 60) & (detections.class_id != 0)]# 解析追踪IDdetections.tracker_id = result.boxes.id.cpu().numpy().astype(int)# 获取每个目标的：追踪ID、类别名称、置信度class_ids = detections.class_id     # 类别IDconfidences = detections.confidence # 置信度tracker_ids = detections.tracker_id # 多目标追踪IDlabels = ['#{} {} {:.1f}'.format(tracker_ids[i], model.names[class_ids[i]], confidences[i]*100) for i in range(len(class_ids))]