视频分析 - 动作识别与跟踪
**前置知识**：需要先掌握 目标检测
前置知识：需要先掌握 目标检测
本文重点：视频分析方法与应用
一、视频基础处理
1.1 视频读写
import cv2
# 读取视频
cap = cv2.VideoCapture('video.mp4')
# 获取视频属性
fps = cap.get(cv2.CAP_PROP_FPS)
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
print(f"FPS: {fps}, 分辨率: {width}x{height}, 总帧数: {frame_count}")
# 创建视频写入器
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out = cv2.VideoWriter('output.mp4', fourcc, fps, (width, height))
# 逐帧处理
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
# 处理帧
processed = process_frame(frame)
# 写入
out.write(processed)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
out.release()
cv2.destroyAllWindows()1.2 帧差法检测运动
def detect_motion(video_path, threshold=25):
"""简单的运动检测"""
cap = cv2.VideoCapture(video_path)
ret, prev_frame = cap.read()
prev_gray = cv2.cvtColor(prev_frame, cv2.COLOR_BGR2GRAY)
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# 计算帧差
diff = cv2.absdiff(prev_gray, gray)
# 二值化
_, thresh = cv2.threshold(diff, threshold, 255, cv2.THRESH_BINARY)
# 膨胀
kernel = np.ones((5, 5), np.uint8)
dilated = cv2.dilate(thresh, kernel, iterations=2)
# 查找轮廓
contours, _ = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
for contour in contours:
if cv2.contourArea(contour) > 1000:
x, y, w, h = cv2.boundingRect(contour)
cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
cv2.imshow('Motion', frame)
prev_gray = gray
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()二、目标跟踪
2.1 传统跟踪算法
# KCF跟踪器
tracker = cv2.TrackerKCF_create()
# 初始化
ret, frame = cap.read()
bbox = (100, 100, 200, 200) # 初始边界框
tracker.init(frame, bbox)
# 跟踪
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
success, bbox = tracker.update(frame)
if success:
x, y, w, h = [int(v) for v in bbox]
cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
else:
cv2.putText(frame, "丢失", (50, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
cv2.imshow('Tracking', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break2.2 多目标跟踪
from ultralytics import YOLO
model = YOLO('yolov8n.pt')
# 视频跟踪
results = model.track(
'video.mp4',
show=True,
tracker='botsort.yaml',
save=True
)
# 获取跟踪信息
for result in results:
boxes = result.boxes
if boxes.id is not None:
track_ids = boxes.id.int().cpu().tolist()
boxes_xyxy = boxes.xyxy.cpu().numpy()
for track_id, box in zip(track_ids, boxes_xyxy):
print(f"ID: {track_id}, 位置: {box}")三、动作识别
3.1 使用预训练模型
import torch
from torchvision.models.video import r3d_18
# 加载预训练模型
model = r3d_18(pretrained=True)
model.eval()
# 准备视频片段 (B, C, T, H, W)
clip = torch.randn(1, 3, 16, 112, 112) # 16帧
# 推理
with torch.no_grad():
output = model(clip)
pred = output.argmax(1)
# Kinetics-400类别标签
print(f"动作类别: {pred.item()}")3.2 使用VideoMAE
from transformers import VideoMAEForVideoClassification, VideoMAEImageProcessor
# 加载模型
model = VideoMAEForVideoClassification.from_pretrained("MCG-NJU/videomae-base-finetuned-kinetics")
processor = VideoMAEImageProcessor.from_pretrained("MCG-NJU/videomae-base-finetuned-kinetics")
# 处理视频
def process_video(video_path, num_frames=16):
frames = []
cap = cv2.VideoCapture(video_path)
# 均匀采样
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
indices = np.linspace(0, total_frames - 1, num_frames, dtype=int)
for idx in indices:
cap.set(cv2.CAP_PROP_POS_FRAMES, idx)
ret, frame = cap.read()
frames.append(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
cap.release()
return frames
# 推理
frames = process_video('action.mp4')
inputs = processor(frames, return_tensors="pt")
with torch.no_grad():
outputs = model(**inputs)
pred = outputs.logits.argmax(-1)
print(f"动作: {model.config.id2label[pred.item()]}")四、行为分析实战
class VideoAnalyzer:
"""视频分析系统"""
def __init__(self):
self.detector = YOLO('yolov8n.pt')
self.track_history = {}
def analyze(self, video_path):
"""分析视频"""
cap = cv2.VideoCapture(video_path)
results_list = []
frame_idx = 0
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
# 检测
results = self.detector.track(frame, persist=True)
# 记录跟踪
if results[0].boxes.id is not None:
boxes = results[0].boxes.xywh.cpu()
track_ids = results[0].boxes.id.int().cpu()
for box, track_id in zip(boxes, track_ids):
x, y, w, h = box
if track_id not in self.track_history:
self.track_history[track_id] = []
self.track_history[track_id].append((float(x), float(y), frame_idx))
frame_idx += 1
cap.release()
return self.track_history
def get_trajectory(self, track_id):
"""获取轨迹"""
return self.track_history.get(track_id, [])
def analyze_behavior(self, track_id):
"""分析行为"""
trajectory = self.get_trajectory(track_id)
if len(trajectory) < 2:
return "轨迹太短"
# 计算移动距离
total_distance = 0
for i in range(1, len(trajectory)):
dx = trajectory[i][0] - trajectory[i-1][0]
dy = trajectory[i][1] - trajectory[i-1][1]
total_distance += (dx**2 + dy**2)**0.5
if total_distance < 100:
return "静止"
elif total_distance < 500:
return "缓慢移动"
else:
return "快速移动"参考资源
返回：计算机视觉基础 最后更新: 2026年4月20日
讨论与反馈