AI视觉实战|人员追踪与重识别:跨摄像头人员轨迹追踪技术详解

技术类型

适用范围

身份保持能力

安防价值

单镜头多目标追踪

固定单路摄像头

出画面立刻丢失ID

只能做局部计数、局部轨迹

跨镜头人员重识别

多路摄像头全域覆盖

跨镜头、跨区域、短时间离场仍保留身份

全域轨迹串联、人员动线分析、事件完整溯源

pythonimport cv2import numpy as npimport torch# ===================== 配置参数 =====================VIDEO_SOURCE = 0CONF_THRESH = 0.5SIMILAR_THRESH = 0.6# 加载检测模型net = cv2.dnn.readNet("yolov3-tiny.cfg","yolov3-tiny.weights")# 简易特征提取（模拟ReID特征）def get_person_feature(crop_img):"""提取行人简易纹理颜色特征向量"""resized = cv2.resize(crop_img,(32,64))mean_color = np.mean(resized,axis=(0,1))gray = cv2.cvtColor(resized,cv2.COLOR_BGR2GRAY)std_texture = np.std(gray)feat = np.concatenate([mean_color,[std_texture]])return feat# 计算特征相似度def feature_similarity(f1,f2):return np.dot(f1,f2)/(np.linalg.norm(f1)*np.linalg.norm(f2))# 全局特征库（模拟跨镜头人员库）person_feature_lib = []cap = cv2.VideoCapture(VIDEO_SOURCE)while True:ret,frame = cap.read()if not ret:breakh,w = frame.shape[:2]frame_show = frame.copy()# 推理检测blob = cv2.dnn.blobFromImage(frame,1/255,(416,416),swapRB=True)net.setInput(blob)outputs = net.forward(net.getUnconnectedOutLayersNames())current_persons = []for out in outputs:for det in out:score = det[5:]cls = np.argmax(score)conf = score[cls]if cls == 0 and conf > CONF_THRESH:cx,cy,bw,bh = det[:4]x = int(cx*w - bw*w/2)y = int(cy*h - bh*h/2)pw = int(bw*w)ph = int(bh*h)# 裁剪行人crop = frame[y:y+ph,x:x+pw]if crop.shape[0]>20 and crop.shape[1]>20:feat = get_person_feature(crop)current_persons.append((x,y,pw,ph,feat))# 重识别匹配逻辑for x,y,pw,ph,feat in current_persons:matched = False# 遍历历史特征库匹配for idx,lib_feat in enumerate(person_feature_lib):sim = feature_similarity(feat,lib_feat)if sim > SIMILAR_THRESH:# 匹配成功，判定为同一人cv2.rectangle(frame_show,(x,y),(x+pw,y+ph),(0,255,0),2)cv2.putText(frame_show,f"ID:{idx+1}",(x,y-10),cv2.FONT_HERSHEY_SIMPLEX,0.6,(0,255,0),2)matched = Truebreakif not matched:# 新人员入库person_feature_lib.append(feat)new_id = len(person_feature_lib)cv2.rectangle(frame_show,(x,y),(x+pw,y+ph),(0,0,255),2)cv2.putText(frame_show,f"NewID:{new_id}",(x,y-10),cv2.FONT_HERSHEY_SIMPLEX,0.6,(0,0,255),2)cv2.putText(frame_show,f"Person Num:{len(current_persons)}",(20,30),cv2.FONT_HERSHEY_SIMPLEX,0.6,(255,255,255),2)cv2.imshow("Person ReID & Track",frame_show)if cv2.waitKey(1)&0xFF == ord('q'):breakcap.release()cv2.destroyAllWindows()