[自用代码] 没活了-用人家的reid做个羽毛球场人员跟踪

二次元刀哥镇楼

简介

先用人家代码实现该功能，以后（刀哥保佑我顺利毕业，否则遥遥无期）大概率要：

• 提高检测器的检测能力（数据训练，新的yolo模型）
• 提高deepsort的分辨能力（低视角，网后人员的辨别）
• 利用mmpose提取球员的骨架
• 球员运动轨迹（仿射变换到二维球场上）
• 羽球的跟踪（双目视觉定位）

具体实现过程：

• 使用query_get.py截取视频中的人物，作为reid的样本
• 加载yolov3.weight权重
• 调整search.py部分参数，运行

1. 截取图片样例，其中包含了网后的图片，但是在deepsort中仍然无法匹配：
   
   
   
   
   
   

2. 其中search.py的代码：

import argparse
import time
from sys import platformfrom models import *
from utils.datasets import *
from utils.utils import *from reid.data import make_data_loader
from reid.data.transforms import build_transforms
from reid.modeling import build_model
from reid.config import cfg as reidCfgdef detect(cfg,data,weights,images='data/samples',  # input folderoutput='output',  # output folderfourcc='mp4v',  # video codecimg_size=416,conf_thres=0.5,nms_thres=0.5,dist_thres=1.0,save_txt=False,save_images=True):# Initializedevice = torch_utils.select_device(force_cpu=False)torch.backends.cudnn.benchmark = False  # set False for reproducible resultsif os.path.exists(output):shutil.rmtree(output)  # delete output folderos.makedirs(output)  # make new output folder############# 行人重识别模型初始化 #############query_loader, num_query = make_data_loader(reidCfg)reidModel = build_model(reidCfg, num_classes=10126)reidModel.load_param(reidCfg.TEST.WEIGHT)reidModel.to(device).eval()query_feats = []query_pids  = []for i, batch in enumerate(query_loader):with torch.no_grad():img, pid, camid = batchimg = img.to(device)feat = reidModel(img)         # 一共2张待查询图片，每张图片特征向量2048 torch.Size([2, 2048])query_feats.append(feat)query_pids.extend(np.asarray(pid))  # extend() 函数用于在列表末尾一次性追加另一个序列中的多个值（用新列表扩展原来的列表）。query_feats = torch.cat(query_feats, dim=0)  # torch.Size([2, 2048])print("The query feature is normalized")query_feats = torch.nn.functional.normalize(query_feats, dim=1, p=2) # 计算出查询图片的特征向量############# 行人检测模型初始化 #############model = Darknet(cfg, img_size)# Load weightsif weights.endswith('.pt'):  # pytorch formatmodel.load_state_dict(torch.load(weights, map_location=device)['model'])else:  # darknet format_ = load_darknet_weights(model, weights)# Eval modemodel.to(device).eval()# Half precisionopt.half = opt.half and device.type != 'cpu'  # half precision only supported on CUDAif opt.half:model.half()# Set Dataloadervid_path, vid_writer = None, Noneif opt.webcam:save_images = Falsedataloader = LoadWebcam(img_size=img_size, half=opt.half)else:dataloader = LoadImages(images, img_size=img_size, half=opt.half)# Get classes and colors# parse_data_cfg(data)['names']:得到类别名称文件路径 names=data/coco.namesclasses = load_classes(parse_data_cfg(data)['names']) # 得到类别名列表: ['person', 'bicycle'...]colors = [[random.randint(0, 255) for _ in range(3)] for _ in range(len(classes))] # 对于每种类别随机使用一种颜色画框# Run inferencet0 = time.time()for i, (path, img, im0, vid_cap) in enumerate(dataloader):t = time.time()# if i < 500 or i % 5 == 0:#     continuesave_path = str(Path(output) / Path(path).name) # 保存的路径# Get detections shape: (3, 416, 320)img = torch.from_numpy(img).unsqueeze(0).to(device) # torch.Size([1, 3, 416, 320])pred, _ = model(img) # 经过处理的网络预测，和原始的det = non_max_suppression(pred.float(), conf_thres, nms_thres)[0] # torch.Size([5, 7])if det is not None and len(det) > 0:# Rescale boxes from 416 to true image size 映射到原图det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()# Print results to screen image 1/3 data\samples\000493.jpg: 288x416 5 persons, Done. (0.869s)print('%gx%g ' % img.shape[2:], end='')  # print image size '288x416'for c in det[:, -1].unique():   # 对图片的所有类进行遍历循环n = (det[:, -1] == c).sum() # 得到了当前类别的个数，也可以用来统计数目if classes[int(c)] == 'person':print('%g %ss' % (n, classes[int(c)]), end=', ') # 打印个数和类别'5 persons'# Draw bounding boxes and labels of detections# (x1y1x2y2, obj_conf, class_conf, class_pred)count = 0gallery_img = []gallery_loc = []for *xyxy, conf, cls_conf, cls in det: # 对于最后的预测框进行遍历# *xyxy: 对于原图来说的左上角右下角坐标: [tensor(349.), tensor(26.), tensor(468.), tensor(341.)]if save_txt:  # Write to filewith open(save_path + '.txt', 'a') as file:file.write(('%g ' * 6 + '\n') % (*xyxy, cls, conf))# Add bbox to the imagelabel = '%s %.2f' % (classes[int(cls)], conf) # 'person 1.00'if classes[int(cls)] == 'person':#plot_one_bo x(xyxy, im0, label=label, color=colors[int(cls)])xmin = int(xyxy[0])ymin = int(xyxy[1])xmax = int(xyxy[2])ymax = int(xyxy[3])w = xmax - xmin # 233h = ymax - ymin # 602# 如果检测到的行人太小了，感觉意义也不大# 这里需要根据实际情况稍微设置下if w*h > 100:gallery_loc.append((xmin, ymin, xmax, ymax))crop_img = im0[ymin:ymax, xmin:xmax] # HWC (602, 233, 3)crop_img = Image.fromarray(cv2.cvtColor(crop_img, cv2.COLOR_BGR2RGB))  # PIL: (233, 602)crop_img = build_transforms(reidCfg)(crop_img).unsqueeze(0)  # torch.Size([1, 3, 256, 128])gallery_img.append(crop_img)if gallery_img:gallery_img = torch.cat(gallery_img, dim=0)  # torch.Size([7, 3, 256, 128])gallery_img = gallery_img.to(device)gallery_feats = reidModel(gallery_img) # torch.Size([7, 2048])print("The gallery feature is normalized")gallery_feats = torch.nn.functional.normalize(gallery_feats, dim=1, p=2)  # 计算出查询图片的特征向量# m: 2# n: 7m, n = query_feats.shape[0], gallery_feats.shape[0]distmat = torch.pow(query_feats, 2).sum(dim=1, keepdim=True).expand(m, n) + \torch.pow(gallery_feats, 2).sum(dim=1, keepdim=True).expand(n, m).t()# out=(beta∗M)+(alpha∗mat1@mat2)# qf^2 + gf^2 - 2 * qf@gf.t()# distmat - 2 * qf@gf.t()# distmat: qf^2 + gf^2# qf: torch.Size([2, 2048])# gf: torch.Size([7, 2048])distmat.addmm_(1, -2, query_feats, gallery_feats.t())# distmat = (qf - gf)^2# distmat = np.array([[1.79536, 2.00926, 0.52790, 1.98851, 2.15138, 1.75929, 1.99410],#                     [1.78843, 1.96036, 0.53674, 1.98929, 1.99490, 1.84878, 1.98575]])distmat = distmat.cpu().numpy()  # <class 'tuple'>: (3, 12)distmat = distmat.sum(axis=0) / len(query_feats) # 平均一下query中同一行人的多个结果index = distmat.argmin()if distmat[index] < dist_thres:print('距离：%s'%distmat[index])plot_one_box(gallery_loc[index], im0, label='(#-(T)-)', color=colors[1])#int(cls)# cv2.imshow('person search', im0)# cv2.waitKey(0)print('Done. (%.3fs)' % (time.time() - t))if opt.webcam:  # Show live webcamcv2.imshow(weights, im0)if save_images:  # Save image with detectionsif dataloader.mode == 'images':cv2.imwrite(save_path, im0)else:if vid_path != save_path:  # new videovid_path = save_pathif isinstance(vid_writer, cv2.VideoWriter):vid_writer.release()  # release previous video writerfps = vid_cap.get(cv2.CAP_PROP_FPS)width = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))height = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*fourcc), fps, (width, height))vid_writer.write(im0)if save_images:print('Results saved to %s' % os.getcwd() + os.sep + output)if platform == 'darwin':  # macosos.system('open ' + output + ' ' + save_path)print('Done. (%.3fs)' % (time.time() - t0))if __name__ == '__main__':parser = argparse.ArgumentParser()parser.add_argument('--cfg', type=str, default='cfg/yolov3.cfg', help="模型配置文件路径")parser.add_argument('--data', type=str, default='data/coco.data', help="数据集配置文件所在路径")parser.add_argument('--weights', type=str, default='weights/yolov3.weights', help='模型权重文件路径')parser.add_argument('--images', type=str, default='data/samples', help='需要进行检测的图片文件夹')parser.add_argument('-q', '--query', default=r'query', help='查询图片的读取路径.')parser.add_argument('--img-size', type=int, default=416, help='输入分辨率大小')parser.add_argument('--conf-thres', type=float, default=0.1, help='物体置信度阈值')parser.add_argument('--nms-thres', type=float, default=0.15, help='NMS阈值')parser.add_argument('--dist_thres', type=float, default=1.0, help='行人图片距离阈值，小于这个距离，就认为是该行人')parser.add_argument('--fourcc', type=str, default='mp4v', help='fourcc output video codec (verify ffmpeg support)')parser.add_argument('--output', type=str, default='output', help='检测后的图片或视频保存的路径')parser.add_argument('--half', default=False, help='是否采用半精度FP16进行推理')parser.add_argument('--webcam', default=False, help='是否使用摄像头进行检测')opt = parser.parse_args()print(opt)with torch.no_grad():detect(opt.cfg,opt.data,opt.weights,images=opt.images,img_size=opt.img_size,conf_thres=opt.conf_thres,nms_thres=opt.nms_thres,dist_thres=opt.dist_thres,fourcc=opt.fourcc,output=opt.output)

输出，可以看到速度还挺快，yolov3能做到几乎实时：

video 1/1 (6335/6341) data\samples\badminton_mot.mp4: 256x416 6 persons, The gallery feature is normalized
Done. (0.055s)
video 1/1 (6336/6341) data\samples\badminton_mot.mp4: 256x416 5 persons, The gallery feature is normalized
Done. (0.047s)
video 1/1 (6337/6341) data\samples\badminton_mot.mp4: 256x416 4 persons, The gallery feature is normalized
Done. (0.050s)
video 1/1 (6338/6341) data\samples\badminton_mot.mp4: 256x416 5 persons, The gallery feature is normalized
Done. (0.045s)
video 1/1 (6339/6341) data\samples\badminton_mot.mp4: 256x416 5 persons, The gallery feature is normalized
Done. (0.046s)
video 1/1 (6340/6341) data\samples\badminton_mot.mp4: 256x416 5 persons, The gallery feature is normalized
Done. (0.052s)
video 1/1 (6341/6341) data\samples\badminton_mot.mp4: 256x416 5 persons, The gallery feature is normalized
Done. (0.045s)
Results saved to E:\sth22_reid\person_search_demo-master\output
Done. (371.999s)

结果动图：