进阶使用技巧
本文档提供了AIToolkit Base的进阶使用技巧，帮助您更高效地使用这个工具包。
多检测器协同工作
在实际应用中，我们可能需要多个检测器协同工作，例如同时进行人脸检测和手势识别：
from aitoolkit_base import Camera, FaceDetector, GestureRecognizer
import cv2
# 同时使用多个检测器
with Camera(0) as camera, FaceDetector() as face_detector, GestureRecognizer() as gesture_detector:
for frame in camera:
# 运行人脸检测
faces = face_detector.run(frame)
# 运行手势识别
gestures = gesture_detector.run(frame)
# 绘制人脸检测结果
result = face_detector.draw(frame, faces)
# 在同一个图像上叠加手势识别结果
result = gesture_detector.draw(result, gestures)
# 显示结果
cv2.imshow("多检测器结果", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
性能优化技巧
降低分辨率
处理高分辨率图像会降低检测速度，可以考虑降低输入图像的分辨率：
from aitoolkit_base import Camera, FaceDetector, ImageUtils
import cv2
with Camera(0) as camera, FaceDetector() as detector:
for frame in camera:
# 降低分辨率以提高速度
small_frame = ImageUtils.resize_image(frame, width=640)
# 运行检测
faces = detector.run(small_frame)
# 绘制结果
result = detector.draw(small_frame, faces)
# 显示结果
cv2.imshow("低分辨率检测", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
减少检测频率
对于实时视频流，可以通过减少检测频率来提高性能：
from aitoolkit_base import Camera, FaceDetector
import cv2
with Camera(0) as camera, FaceDetector() as detector:
frame_count = 0
last_faces = None
for frame in camera:
# 每3帧检测一次
if frame_count % 3 == 0:
faces = detector.run(frame)
last_faces = faces
else:
faces = last_faces
# 绘制结果
if faces:
result = detector.draw(frame, faces)
else:
result = frame
# 显示结果
cv2.imshow("减少检测频率", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
frame_count += 1
调整检测参数
适当调整检测参数也可以提高性能：
from aitoolkit_base import Camera, FaceDetector
import cv2
# 通过降低检测置信度阈值和最大检测数量来提高性能
with Camera(0) as camera, FaceDetector(min_detection_confidence=0.3, max_num_faces=1) as detector:
for frame in camera:
faces = detector.run(frame)
result = detector.draw(frame, faces)
cv2.imshow("优化参数", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
错误处理与异常恢复
在实际应用中，良好的错误处理非常重要：
from aitoolkit_base import Camera, FaceDetector
import cv2
import time
# 启动摄像头和检测器
try:
camera = Camera(0, auto_reconnect=True)
detector = FaceDetector()
while True:
try:
# 读取帧
ret, frame = camera.read()
if not ret:
print("无法读取帧，等待...")
time.sleep(0.5)
continue
# 运行检测
try:
faces = detector.run(frame)
result = detector.draw(frame, faces)
except Exception as e:
print(f"检测失败: {str(e)}")
result = frame
# 显示结果
cv2.imshow("错误处理", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
except Exception as e:
print(f"处理异常: {str(e)}")
time.sleep(1) # 出错时暂停一秒再继续
except Exception as e:
print(f"严重错误: {str(e)}")
finally:
# 确保资源被释放
if 'camera' in locals():
camera.release()
if 'detector' in locals():
detector.close()
cv2.destroyAllWindows()
自定义检测结果处理
您可以自定义如何处理检测结果：
from aitoolkit_base import Camera, FaceDetector
import cv2
import time
# 自定义人脸信息处理
def process_face_info(faces):
if not faces:
return "未检测到人脸"
result = []
for i, face in enumerate(faces):
confidence = face["score"] * 100
box = face["bounding_box"]
size = box["width"] * box["height"]
result.append(f"人脸{i+1}: 置信度={confidence:.1f}%, 大小={size}像素")
return "\n".join(result)
# 使用自定义处理
with Camera(0) as camera, FaceDetector() as detector:
for frame in camera:
# 运行检测
faces = detector.run(frame)
# 绘制基本结果
result = detector.draw(frame, faces)
# 处理并显示自定义信息
info = process_face_info(faces)
y_offset = 30
for line in info.split("\n"):
cv2.putText(result, line, (10, y_offset),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2)
y_offset += 30
# 显示结果
cv2.imshow("自定义处理", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
录制视频
将检测结果保存为视频文件：
from aitoolkit_base import Camera, FaceDetector
import cv2
# 设置视频录制参数
camera = Camera(0, width=1280, height=720)
detector = FaceDetector()
# 创建视频写入器
fourcc = cv2.VideoWriter_fourcc(*'XVID')
out = cv2.VideoWriter('output.avi', fourcc, 20.0, (1280, 720))
try:
for frame in camera:
# 运行检测
faces = detector.run(frame)
result = detector.draw(frame, faces)
# 显示结果
cv2.imshow("录制中", result)
# 写入视频
out.write(result)
# 按q退出
if cv2.waitKey(1) & 0xFF == ord('q'):
break
finally:
# 释放资源
camera.release()
detector.close()
out.release()
cv2.destroyAllWindows()
使用异步模式
异步模式可以提高处理效率：
from aitoolkit_base import Camera, FaceDetector
import cv2
import threading
import queue
import time
# 创建队列用于存储处理结果
result_queue = queue.Queue(maxsize=5)
# 异步处理线程
def process_frames(camera, detector, result_queue):
for frame in camera:
# 运行检测
faces = detector.run(frame)
result = detector.draw(frame, faces)
# 将结果放入队列，如果队列满则等待
try:
result_queue.put((result, time.time()), block=False)
except queue.Full:
# 队列满，丢弃当前帧
pass
# 主程序
with Camera(0) as camera, FaceDetector() as detector:
# 启动处理线程
processing_thread = threading.Thread(
target=process_frames,
args=(camera, detector, result_queue)
)
processing_thread.daemon = True
processing_thread.start()
# 主线程负责显示
while True:
try:
# 获取处理结果，超时1秒
result, timestamp = result_queue.get(timeout=1)
# 计算延迟
delay = time.time() - timestamp
# 显示结果和延迟
cv2.putText(result, f"Delay: {delay*1000:.1f}ms", (10, 30),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
cv2.imshow("异步处理", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
except queue.Empty:
continue
组合多种机器视觉功能
将不同的机器视觉功能组合使用：
from aitoolkit_base import Camera, FaceDetector, HandLandmarker
import cv2
import numpy as np
# 在同一帧上组合多种视觉功能
with Camera(0) as camera, FaceDetector() as face_detector, HandLandmarker() as hand_detector:
for frame in camera:
# 创建一个原始帧的副本
display = frame.copy()
# 添加人脸检测
faces = face_detector.run(frame)
if faces:
display = face_detector.draw(display, faces)
# 添加手部关键点检测
hands = hand_detector.run(frame)
if hands:
display = hand_detector.draw(display, hands)
# 显示结果
cv2.imshow("视觉功能组合", display)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
以上就是AIToolkit Base的一些进阶使用技巧，通过这些技巧，您可以更高效地使用这个工具包来实现复杂的应用。