Spaces:
Running
Running
Upload 6 files
Browse files- app.py +177 -0
- detectMotion.py +60 -0
- openh264-1.8.0-win64.dll +0 -0
- requirements.txt +4 -0
- sort.py +330 -0
- yolov8n.pt +3 -0
app.py
ADDED
@@ -0,0 +1,177 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from ultralytics import YOLO #行人识别,采用YoloV8模型
|
2 |
+
import math #用于四舍五入取整
|
3 |
+
import cv2 #opencv图像处理库
|
4 |
+
import cvzone #在图像上绘画
|
5 |
+
#import numpy as np #忘记取消注释了...
|
6 |
+
import gradio as gr #GUI界面
|
7 |
+
from sort import * #运动检测,采用sort算法
|
8 |
+
#from deep_sort_realtime.deepsort_tracker import DeepSort #不会用QAQ
|
9 |
+
import tempfile #创建输出临时文件夹
|
10 |
+
import os
|
11 |
+
from detectMotion import * #单独的运动检测
|
12 |
+
|
13 |
+
#导入YoloV8模型,第一次使用会下载模型到当前文件夹当中
|
14 |
+
model=YOLO("yolov8n.pt")
|
15 |
+
|
16 |
+
#YoloV8标签数据,本次项目只使用了'person'
|
17 |
+
classNames=['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light',
|
18 |
+
'fire hydrant',
|
19 |
+
'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra',
|
20 |
+
'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball',
|
21 |
+
'kite',
|
22 |
+
'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork',
|
23 |
+
'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza',
|
24 |
+
'donut',
|
25 |
+
'cake', 'chair', 'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote',
|
26 |
+
'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase',
|
27 |
+
'scissors',
|
28 |
+
'teddy bear', 'hair drier', 'toothbrush']
|
29 |
+
|
30 |
+
#运动检测算法参数
|
31 |
+
tracker=Sort(max_age=20,min_hits=3,iou_threshold=0.3)
|
32 |
+
|
33 |
+
# 彩色图像进行自适应直方图均衡化
|
34 |
+
def hisEqulColor(img):
|
35 |
+
## 将RGB图像转换到YCrCb空间中
|
36 |
+
ycrcb = cv2.cvtColor(img, cv2.COLOR_BGR2YCR_CB)
|
37 |
+
# 将YCrCb图像通道分离
|
38 |
+
channels = cv2.split(ycrcb)
|
39 |
+
# 以下代码详细注释见官网:
|
40 |
+
# https://docs.opencv.org/4.1.0/d5/daf/tutorial_py_histogram_equalization.html
|
41 |
+
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
|
42 |
+
clahe.apply(channels[0], channels[0])
|
43 |
+
cv2.merge(channels, ycrcb)
|
44 |
+
cv2.cvtColor(ycrcb, cv2.COLOR_YCR_CB2BGR, img)
|
45 |
+
return img
|
46 |
+
|
47 |
+
#添加高斯噪声,并使用中值滤波降噪
|
48 |
+
def AddGaussNoise(img,sigma):
|
49 |
+
gauss=np.random.normal(0,sigma,img.shape)
|
50 |
+
img=np.uint8(img + gauss)#将高斯噪声与原始图像叠加
|
51 |
+
img=cv2.medianBlur(img,5)
|
52 |
+
return img
|
53 |
+
|
54 |
+
#图像处理
|
55 |
+
def processImg(img,sigma):
|
56 |
+
img=cv2.cvtColor(img,cv2.COLOR_RGB2BGR)
|
57 |
+
res1 = AddGaussNoise(img,sigma)
|
58 |
+
res1 = hisEqulColor(res1)
|
59 |
+
res1=cv2.cvtColor(res1,cv2.COLOR_BGR2RGB)
|
60 |
+
return res1
|
61 |
+
|
62 |
+
#视频处理
|
63 |
+
def processVideo(inputPath,codec):
|
64 |
+
number_of_people=0
|
65 |
+
cap = cv2.VideoCapture(inputPath)#从inputPath读入视频
|
66 |
+
fps = cap.get(cv2.CAP_PROP_FPS) #获取视频的帧率
|
67 |
+
size = (int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)),
|
68 |
+
int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)))#获取视频的大小
|
69 |
+
output_viedo = cv2.VideoWriter()#初始化视频写入
|
70 |
+
outputPath=tempfile.mkdtemp()#创建输出视频的临时文件夹的路径
|
71 |
+
|
72 |
+
#输出格式
|
73 |
+
if codec == "mp4":
|
74 |
+
fourcc = cv2.VideoWriter_fourcc('a','v','c','1')#视频编码:h264,只有h264格式的mp4文件才能在浏览器直接播放
|
75 |
+
video_save_path = os.path.join(outputPath,"output.mp4")#创建输出视频路径
|
76 |
+
elif codec == "avi":
|
77 |
+
fourcc = cv2.VideoWriter_fourcc('h','2','6','4')#视频编码:h264,只能保存为avi格式且不能在浏览器直接播放
|
78 |
+
video_save_path = os.path.join(outputPath,"output.avi")#创建输出视频路径
|
79 |
+
elif codec == "mkv":
|
80 |
+
fourcc = cv2.VideoWriter_fourcc('X','V','I','D')#视频编码:XVID,此编码不需要openh264-1.8.0-win64.dll
|
81 |
+
video_save_path = os.path.join(outputPath,"output.mkv")#创建输出视频路径
|
82 |
+
|
83 |
+
output_viedo.open(video_save_path , fourcc, fps, size, True)
|
84 |
+
#对每一帧图片进行读取和处理
|
85 |
+
while True:
|
86 |
+
ret, img = cap.read()#将每一帧图片读取到img当中
|
87 |
+
results=model(img,stream=True)#使用YoloV8模型进行推理,stream=True提高流式处理速度
|
88 |
+
detections=np.empty((0, 5))#初始化运动检测
|
89 |
+
if not(ret):#当视频全部读完,ret返回false,终止循环,视频帧读取和写入结束
|
90 |
+
break
|
91 |
+
img = hisEqulColor(img)#视频增强
|
92 |
+
#读取推理的数据
|
93 |
+
for r in results:
|
94 |
+
boxes=r.boxes
|
95 |
+
for box in boxes:
|
96 |
+
#读取每一帧识别出的边界信息,并显示
|
97 |
+
x1,y1,x2,y2=box.xyxy[0]
|
98 |
+
x1,y1,x2,y2=int(x1),int(y1),int(x2),int(y2)#将tensor类型转变为整型
|
99 |
+
conf=math.ceil(box.conf[0]*100)/100#对conf取2位小数
|
100 |
+
cls=int(box.cls[0])#获取物体类别标签
|
101 |
+
#当标签为人,且可信度大于0.3的时候,将人标识出来
|
102 |
+
if cls==0 and conf > 0.3:
|
103 |
+
cv2.rectangle(img,(x1,y1),(x2,y2),(255,0,255),3)
|
104 |
+
print(conf)
|
105 |
+
#cvzone.putTextRect(img,f'{classNames[cls]}{conf}',(max(0,x1),max(30,y1)),scale=0.7,thickness=1)
|
106 |
+
currentArray=np.array([x1,y1,x2,y2,conf])
|
107 |
+
detections=np.vstack((detections,currentArray))#按行堆叠数据
|
108 |
+
#运动检测
|
109 |
+
resultsTracker=tracker.update(detections)
|
110 |
+
for result in resultsTracker:
|
111 |
+
x1,y1,x2,y2,Id=result
|
112 |
+
number_of_people=max(str(int(Id)),str(number_of_people))
|
113 |
+
x1,y1,x2,y2=int(x1),int(y1),int(x2),int(y2)#将浮点数转变为整型
|
114 |
+
#print(result)
|
115 |
+
cvzone.putTextRect(img,f'{int(Id)}',(max(0,x1),max(30,y1)),scale=0.7,thickness=1)
|
116 |
+
#image_np = np.squeeze(img.render())#用np.squeeze将输出结果降维
|
117 |
+
output_viedo.write(img)#将处理后的图像写入视频
|
118 |
+
output_viedo.release()#释放
|
119 |
+
cap.release()#释放
|
120 |
+
return video_save_path,video_save_path
|
121 |
+
|
122 |
+
|
123 |
+
|
124 |
+
#WebUi图形界面(interface)
|
125 |
+
#demo = gr.Interface(
|
126 |
+
# fn=processVideo,
|
127 |
+
# inputs=["text","text"],
|
128 |
+
# outputs=["text"],
|
129 |
+
# examples=[["D:\\[WPF]JJDown\\Download\\walker.mp4","C:\\Users\\sino\\Downloads\\output.mkv"],["C:\\Users\\sino\\Videos\\test.mp4","C:\\Users\\sino\\Downloads\\output.mkv"]],
|
130 |
+
# title="运动检测与行人跟踪",
|
131 |
+
# description="请输入绝对路径"
|
132 |
+
#)
|
133 |
+
|
134 |
+
#WebUi图形界面(block)
|
135 |
+
with gr.Blocks() as demo:
|
136 |
+
gr.Markdown("""
|
137 |
+
# 运动检测与行人跟踪
|
138 |
+
基于opencv + yoloV8 + sort,请输入绝对路径
|
139 |
+
""")
|
140 |
+
with gr.Tab("视频识别"):
|
141 |
+
with gr.Row():
|
142 |
+
with gr.Column():
|
143 |
+
text_inputPath = gr.Video()
|
144 |
+
codec = gr.Radio(["mp4","avi","mkv"], label="输出视频格式")
|
145 |
+
videoProcess_button = gr.Button("处理")
|
146 |
+
with gr.Column():
|
147 |
+
text_output = gr.Video()
|
148 |
+
text_output_path = gr.Text(label="输出路径")
|
149 |
+
with gr.Tab("图像增强"):
|
150 |
+
with gr.Row():
|
151 |
+
with gr.Column():
|
152 |
+
image_input = gr.Image()
|
153 |
+
image_sigma = gr.Slider(0,40,label="高斯噪声sigma")
|
154 |
+
image_output = gr.Image()
|
155 |
+
image_button = gr.Button("处理")
|
156 |
+
with gr.Tab("运动检测"):
|
157 |
+
with gr.Column():
|
158 |
+
with gr.Row():
|
159 |
+
with gr.Column():
|
160 |
+
motion_inputPath = gr.Video()
|
161 |
+
motionProcess_button = gr.Button("处理")
|
162 |
+
motion_output_frame = gr.Video()
|
163 |
+
motion_output_fmask = gr.Video()
|
164 |
+
frame_output_path = gr.Text(label="frame输出路径")
|
165 |
+
fmask_output_path = gr.Text(label="mask输出路径")
|
166 |
+
with gr.Accordion("算法:"):
|
167 |
+
gr.Markdown("高斯混合模型(GMM)")
|
168 |
+
|
169 |
+
|
170 |
+
|
171 |
+
videoProcess_button.click(processVideo, inputs=[text_inputPath,codec], outputs=[text_output,text_output_path])
|
172 |
+
image_button.click(processImg, inputs=[image_input,image_sigma], outputs=image_output)
|
173 |
+
motionProcess_button.click(motionDetection, inputs=[motion_inputPath], outputs=[motion_output_frame,motion_output_fmask,
|
174 |
+
frame_output_path,fmask_output_path])
|
175 |
+
|
176 |
+
demo.queue()#当有多个请求时,排队
|
177 |
+
demo.launch()#生成内网链接,如需要公网链接,括号内输入share=True
|
detectMotion.py
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#修改自https://blog.csdn.net/qq_29367075/article/details/122933407
|
2 |
+
import cv2
|
3 |
+
import numpy as np
|
4 |
+
import tempfile
|
5 |
+
import os
|
6 |
+
|
7 |
+
kernel1 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5))
|
8 |
+
mog = cv2.createBackgroundSubtractorMOG2() # 创建混合高斯模型来用于北京建模
|
9 |
+
|
10 |
+
def motionDetection(inputPath):
|
11 |
+
cap = cv2.VideoCapture(inputPath)#从inputPath读入视频
|
12 |
+
fps = cap.get(cv2.CAP_PROP_FPS) #获取视频的帧率
|
13 |
+
size = (int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)),
|
14 |
+
int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)))#获取视频的大小
|
15 |
+
output_viedo_frame = cv2.VideoWriter()#初始化视频写入
|
16 |
+
output_viedo_fmask = cv2.VideoWriter()#初始化视频写入
|
17 |
+
outputPath=tempfile.mkdtemp()#创建输出视频的临时文件夹的路径
|
18 |
+
fourcc = cv2.VideoWriter_fourcc('a','v','c','1')#视频编码:h264,只有h264格式的mp4文件才能在浏览器直接播放
|
19 |
+
video_save_path_frame = os.path.join(outputPath,"frame.mp4")#创建输出视频路径
|
20 |
+
video_save_path_fmask = os.path.join(outputPath,"fmask.mp4")#创建输出视频路径
|
21 |
+
output_viedo_frame.open(video_save_path_frame , fourcc, fps, size, True)
|
22 |
+
output_viedo_fmask.open(video_save_path_fmask , fourcc, fps, size, True)
|
23 |
+
#对每一帧图片进行读取和处理
|
24 |
+
while True:
|
25 |
+
ret, frame = cap.read()#将每一帧图片读取到img当中
|
26 |
+
if frame is None:
|
27 |
+
print("camera is over...")
|
28 |
+
break
|
29 |
+
|
30 |
+
fmask = mog.apply(frame) # 判断哪些是前景和背景
|
31 |
+
|
32 |
+
|
33 |
+
MORPH_OPEN_1 = cv2.morphologyEx(fmask, cv2.MORPH_OPEN, kernel1) # 开运算,去除噪声和毛刺
|
34 |
+
|
35 |
+
contours, _ = cv2.findContours(fmask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # 只检测外边框
|
36 |
+
|
37 |
+
for cont in contours:
|
38 |
+
# 计算各个轮廓的面积
|
39 |
+
len = cv2.arcLength(cont, True)
|
40 |
+
if len > 300: # 去除一些小的噪声点
|
41 |
+
# 找到一个轮廓
|
42 |
+
x,y,w,h = cv2.boundingRect(cont)
|
43 |
+
# 画出这个矩形
|
44 |
+
cv2.rectangle(frame, (x,y), (x+w, y+h), color=(0,255,0), thickness=3)
|
45 |
+
fmask=cv2.cvtColor(fmask,cv2.COLOR_BGR2RGB)
|
46 |
+
#print(fmask)
|
47 |
+
#image_np = np.squeeze(img.render())#用np.squeeze将输出结果降维
|
48 |
+
output_viedo_frame.write(frame)#将处理后的图像写入视频
|
49 |
+
output_viedo_fmask.write(fmask)#将处理后的图像写入视频
|
50 |
+
output_viedo_frame.release()#释放
|
51 |
+
output_viedo_fmask.release()#释放
|
52 |
+
cap.release()#释放
|
53 |
+
return video_save_path_frame,video_save_path_fmask,video_save_path_frame,video_save_path_fmask
|
54 |
+
|
55 |
+
|
56 |
+
|
57 |
+
|
58 |
+
|
59 |
+
|
60 |
+
|
openh264-1.8.0-win64.dll
ADDED
Binary file (825 kB). View file
|
|
requirements.txt
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
ultralytics
|
2 |
+
opencv-python
|
3 |
+
cvzone
|
4 |
+
torch
|
sort.py
ADDED
@@ -0,0 +1,330 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
SORT: A Simple, Online and Realtime Tracker
|
3 |
+
Copyright (C) 2016-2020 Alex Bewley alex@bewley.ai
|
4 |
+
|
5 |
+
This program is free software: you can redistribute it and/or modify
|
6 |
+
it under the terms of the GNU General Public License as published by
|
7 |
+
the Free Software Foundation, either version 3 of the License, or
|
8 |
+
(at your option) any later version.
|
9 |
+
|
10 |
+
This program is distributed in the hope that it will be useful,
|
11 |
+
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
12 |
+
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
13 |
+
GNU General Public License for more details.
|
14 |
+
|
15 |
+
You should have received a copy of the GNU General Public License
|
16 |
+
along with this program. If not, see <http://www.gnu.org/licenses/>.
|
17 |
+
"""
|
18 |
+
from __future__ import print_function
|
19 |
+
|
20 |
+
import os
|
21 |
+
import numpy as np
|
22 |
+
import matplotlib
|
23 |
+
matplotlib.use('TkAgg')
|
24 |
+
import matplotlib.pyplot as plt
|
25 |
+
import matplotlib.patches as patches
|
26 |
+
from skimage import io
|
27 |
+
|
28 |
+
import glob
|
29 |
+
import time
|
30 |
+
import argparse
|
31 |
+
from filterpy.kalman import KalmanFilter
|
32 |
+
|
33 |
+
np.random.seed(0)
|
34 |
+
|
35 |
+
|
36 |
+
def linear_assignment(cost_matrix):
|
37 |
+
try:
|
38 |
+
import lap
|
39 |
+
_, x, y = lap.lapjv(cost_matrix, extend_cost=True)
|
40 |
+
return np.array([[y[i],i] for i in x if i >= 0]) #
|
41 |
+
except ImportError:
|
42 |
+
from scipy.optimize import linear_sum_assignment
|
43 |
+
x, y = linear_sum_assignment(cost_matrix)
|
44 |
+
return np.array(list(zip(x, y)))
|
45 |
+
|
46 |
+
|
47 |
+
def iou_batch(bb_test, bb_gt):
|
48 |
+
"""
|
49 |
+
From SORT: Computes IOU between two bboxes in the form [x1,y1,x2,y2]
|
50 |
+
"""
|
51 |
+
bb_gt = np.expand_dims(bb_gt, 0)
|
52 |
+
bb_test = np.expand_dims(bb_test, 1)
|
53 |
+
|
54 |
+
xx1 = np.maximum(bb_test[..., 0], bb_gt[..., 0])
|
55 |
+
yy1 = np.maximum(bb_test[..., 1], bb_gt[..., 1])
|
56 |
+
xx2 = np.minimum(bb_test[..., 2], bb_gt[..., 2])
|
57 |
+
yy2 = np.minimum(bb_test[..., 3], bb_gt[..., 3])
|
58 |
+
w = np.maximum(0., xx2 - xx1)
|
59 |
+
h = np.maximum(0., yy2 - yy1)
|
60 |
+
wh = w * h
|
61 |
+
o = wh / ((bb_test[..., 2] - bb_test[..., 0]) * (bb_test[..., 3] - bb_test[..., 1])
|
62 |
+
+ (bb_gt[..., 2] - bb_gt[..., 0]) * (bb_gt[..., 3] - bb_gt[..., 1]) - wh)
|
63 |
+
return(o)
|
64 |
+
|
65 |
+
|
66 |
+
def convert_bbox_to_z(bbox):
|
67 |
+
"""
|
68 |
+
Takes a bounding box in the form [x1,y1,x2,y2] and returns z in the form
|
69 |
+
[x,y,s,r] where x,y is the centre of the box and s is the scale/area and r is
|
70 |
+
the aspect ratio
|
71 |
+
"""
|
72 |
+
w = bbox[2] - bbox[0]
|
73 |
+
h = bbox[3] - bbox[1]
|
74 |
+
x = bbox[0] + w/2.
|
75 |
+
y = bbox[1] + h/2.
|
76 |
+
s = w * h #scale is just area
|
77 |
+
r = w / float(h)
|
78 |
+
return np.array([x, y, s, r]).reshape((4, 1))
|
79 |
+
|
80 |
+
|
81 |
+
def convert_x_to_bbox(x,score=None):
|
82 |
+
"""
|
83 |
+
Takes a bounding box in the centre form [x,y,s,r] and returns it in the form
|
84 |
+
[x1,y1,x2,y2] where x1,y1 is the top left and x2,y2 is the bottom right
|
85 |
+
"""
|
86 |
+
w = np.sqrt(x[2] * x[3])
|
87 |
+
h = x[2] / w
|
88 |
+
if(score==None):
|
89 |
+
return np.array([x[0]-w/2.,x[1]-h/2.,x[0]+w/2.,x[1]+h/2.]).reshape((1,4))
|
90 |
+
else:
|
91 |
+
return np.array([x[0]-w/2.,x[1]-h/2.,x[0]+w/2.,x[1]+h/2.,score]).reshape((1,5))
|
92 |
+
|
93 |
+
|
94 |
+
class KalmanBoxTracker(object):
|
95 |
+
"""
|
96 |
+
This class represents the internal state of individual tracked objects observed as bbox.
|
97 |
+
"""
|
98 |
+
count = 0
|
99 |
+
def __init__(self,bbox):
|
100 |
+
"""
|
101 |
+
Initialises a tracker using initial bounding box.
|
102 |
+
"""
|
103 |
+
#define constant velocity model
|
104 |
+
self.kf = KalmanFilter(dim_x=7, dim_z=4)
|
105 |
+
self.kf.F = np.array([[1,0,0,0,1,0,0],[0,1,0,0,0,1,0],[0,0,1,0,0,0,1],[0,0,0,1,0,0,0], [0,0,0,0,1,0,0],[0,0,0,0,0,1,0],[0,0,0,0,0,0,1]])
|
106 |
+
self.kf.H = np.array([[1,0,0,0,0,0,0],[0,1,0,0,0,0,0],[0,0,1,0,0,0,0],[0,0,0,1,0,0,0]])
|
107 |
+
|
108 |
+
self.kf.R[2:,2:] *= 10.
|
109 |
+
self.kf.P[4:,4:] *= 1000. #give high uncertainty to the unobservable initial velocities
|
110 |
+
self.kf.P *= 10.
|
111 |
+
self.kf.Q[-1,-1] *= 0.01
|
112 |
+
self.kf.Q[4:,4:] *= 0.01
|
113 |
+
|
114 |
+
self.kf.x[:4] = convert_bbox_to_z(bbox)
|
115 |
+
self.time_since_update = 0
|
116 |
+
self.id = KalmanBoxTracker.count
|
117 |
+
KalmanBoxTracker.count += 1
|
118 |
+
self.history = []
|
119 |
+
self.hits = 0
|
120 |
+
self.hit_streak = 0
|
121 |
+
self.age = 0
|
122 |
+
|
123 |
+
def update(self,bbox):
|
124 |
+
"""
|
125 |
+
Updates the state vector with observed bbox.
|
126 |
+
"""
|
127 |
+
self.time_since_update = 0
|
128 |
+
self.history = []
|
129 |
+
self.hits += 1
|
130 |
+
self.hit_streak += 1
|
131 |
+
self.kf.update(convert_bbox_to_z(bbox))
|
132 |
+
|
133 |
+
def predict(self):
|
134 |
+
"""
|
135 |
+
Advances the state vector and returns the predicted bounding box estimate.
|
136 |
+
"""
|
137 |
+
if((self.kf.x[6]+self.kf.x[2])<=0):
|
138 |
+
self.kf.x[6] *= 0.0
|
139 |
+
self.kf.predict()
|
140 |
+
self.age += 1
|
141 |
+
if(self.time_since_update>0):
|
142 |
+
self.hit_streak = 0
|
143 |
+
self.time_since_update += 1
|
144 |
+
self.history.append(convert_x_to_bbox(self.kf.x))
|
145 |
+
return self.history[-1]
|
146 |
+
|
147 |
+
def get_state(self):
|
148 |
+
"""
|
149 |
+
Returns the current bounding box estimate.
|
150 |
+
"""
|
151 |
+
return convert_x_to_bbox(self.kf.x)
|
152 |
+
|
153 |
+
|
154 |
+
def associate_detections_to_trackers(detections,trackers,iou_threshold = 0.3):
|
155 |
+
"""
|
156 |
+
Assigns detections to tracked object (both represented as bounding boxes)
|
157 |
+
|
158 |
+
Returns 3 lists of matches, unmatched_detections and unmatched_trackers
|
159 |
+
"""
|
160 |
+
if(len(trackers)==0):
|
161 |
+
return np.empty((0,2),dtype=int), np.arange(len(detections)), np.empty((0,5),dtype=int)
|
162 |
+
|
163 |
+
iou_matrix = iou_batch(detections, trackers)
|
164 |
+
|
165 |
+
if min(iou_matrix.shape) > 0:
|
166 |
+
a = (iou_matrix > iou_threshold).astype(np.int32)
|
167 |
+
if a.sum(1).max() == 1 and a.sum(0).max() == 1:
|
168 |
+
matched_indices = np.stack(np.where(a), axis=1)
|
169 |
+
else:
|
170 |
+
matched_indices = linear_assignment(-iou_matrix)
|
171 |
+
else:
|
172 |
+
matched_indices = np.empty(shape=(0,2))
|
173 |
+
|
174 |
+
unmatched_detections = []
|
175 |
+
for d, det in enumerate(detections):
|
176 |
+
if(d not in matched_indices[:,0]):
|
177 |
+
unmatched_detections.append(d)
|
178 |
+
unmatched_trackers = []
|
179 |
+
for t, trk in enumerate(trackers):
|
180 |
+
if(t not in matched_indices[:,1]):
|
181 |
+
unmatched_trackers.append(t)
|
182 |
+
|
183 |
+
#filter out matched with low IOU
|
184 |
+
matches = []
|
185 |
+
for m in matched_indices:
|
186 |
+
if(iou_matrix[m[0], m[1]]<iou_threshold):
|
187 |
+
unmatched_detections.append(m[0])
|
188 |
+
unmatched_trackers.append(m[1])
|
189 |
+
else:
|
190 |
+
matches.append(m.reshape(1,2))
|
191 |
+
if(len(matches)==0):
|
192 |
+
matches = np.empty((0,2),dtype=int)
|
193 |
+
else:
|
194 |
+
matches = np.concatenate(matches,axis=0)
|
195 |
+
|
196 |
+
return matches, np.array(unmatched_detections), np.array(unmatched_trackers)
|
197 |
+
|
198 |
+
|
199 |
+
class Sort(object):
|
200 |
+
def __init__(self, max_age=1, min_hits=3, iou_threshold=0.3):
|
201 |
+
"""
|
202 |
+
Sets key parameters for SORT
|
203 |
+
"""
|
204 |
+
self.max_age = max_age
|
205 |
+
self.min_hits = min_hits
|
206 |
+
self.iou_threshold = iou_threshold
|
207 |
+
self.trackers = []
|
208 |
+
self.frame_count = 0
|
209 |
+
|
210 |
+
def update(self, dets=np.empty((0, 5))):
|
211 |
+
"""
|
212 |
+
Params:
|
213 |
+
dets - a numpy array of detections in the format [[x1,y1,x2,y2,score],[x1,y1,x2,y2,score],...]
|
214 |
+
Requires: this method must be called once for each frame even with empty detections (use np.empty((0, 5)) for frames without detections).
|
215 |
+
Returns the a similar array, where the last column is the object ID.
|
216 |
+
|
217 |
+
NOTE: The number of objects returned may differ from the number of detections provided.
|
218 |
+
"""
|
219 |
+
self.frame_count += 1
|
220 |
+
# get predicted locations from existing trackers.
|
221 |
+
trks = np.zeros((len(self.trackers), 5))
|
222 |
+
to_del = []
|
223 |
+
ret = []
|
224 |
+
for t, trk in enumerate(trks):
|
225 |
+
pos = self.trackers[t].predict()[0]
|
226 |
+
trk[:] = [pos[0], pos[1], pos[2], pos[3], 0]
|
227 |
+
if np.any(np.isnan(pos)):
|
228 |
+
to_del.append(t)
|
229 |
+
trks = np.ma.compress_rows(np.ma.masked_invalid(trks))
|
230 |
+
for t in reversed(to_del):
|
231 |
+
self.trackers.pop(t)
|
232 |
+
matched, unmatched_dets, unmatched_trks = associate_detections_to_trackers(dets,trks, self.iou_threshold)
|
233 |
+
|
234 |
+
# update matched trackers with assigned detections
|
235 |
+
for m in matched:
|
236 |
+
self.trackers[m[1]].update(dets[m[0], :])
|
237 |
+
|
238 |
+
# create and initialise new trackers for unmatched detections
|
239 |
+
for i in unmatched_dets:
|
240 |
+
trk = KalmanBoxTracker(dets[i,:])
|
241 |
+
self.trackers.append(trk)
|
242 |
+
i = len(self.trackers)
|
243 |
+
for trk in reversed(self.trackers):
|
244 |
+
d = trk.get_state()[0]
|
245 |
+
if (trk.time_since_update < 1) and (trk.hit_streak >= self.min_hits or self.frame_count <= self.min_hits):
|
246 |
+
ret.append(np.concatenate((d,[trk.id+1])).reshape(1,-1)) # +1 as MOT benchmark requires positive
|
247 |
+
i -= 1
|
248 |
+
# remove dead tracklet
|
249 |
+
if(trk.time_since_update > self.max_age):
|
250 |
+
self.trackers.pop(i)
|
251 |
+
if(len(ret)>0):
|
252 |
+
return np.concatenate(ret)
|
253 |
+
return np.empty((0,5))
|
254 |
+
|
255 |
+
def parse_args():
|
256 |
+
"""Parse input arguments."""
|
257 |
+
parser = argparse.ArgumentParser(description='SORT demo')
|
258 |
+
parser.add_argument('--display', dest='display', help='Display online tracker output (slow) [False]',action='store_true')
|
259 |
+
parser.add_argument("--seq_path", help="Path to detections.", type=str, default='data')
|
260 |
+
parser.add_argument("--phase", help="Subdirectory in seq_path.", type=str, default='train')
|
261 |
+
parser.add_argument("--max_age",
|
262 |
+
help="Maximum number of frames to keep alive a track without associated detections.",
|
263 |
+
type=int, default=1)
|
264 |
+
parser.add_argument("--min_hits",
|
265 |
+
help="Minimum number of associated detections before track is initialised.",
|
266 |
+
type=int, default=3)
|
267 |
+
parser.add_argument("--iou_threshold", help="Minimum IOU for match.", type=float, default=0.3)
|
268 |
+
args = parser.parse_args()
|
269 |
+
return args
|
270 |
+
|
271 |
+
if __name__ == '__main__':
|
272 |
+
# all train
|
273 |
+
args = parse_args()
|
274 |
+
display = args.display
|
275 |
+
phase = args.phase
|
276 |
+
total_time = 0.0
|
277 |
+
total_frames = 0
|
278 |
+
colours = np.random.rand(32, 3) #used only for display
|
279 |
+
if(display):
|
280 |
+
if not os.path.exists('mot_benchmark'):
|
281 |
+
print('\n\tERROR: mot_benchmark link not found!\n\n Create a symbolic link to the MOT benchmark\n (https://motchallenge.net/data/2D_MOT_2015/#download). E.g.:\n\n $ ln -s /path/to/MOT2015_challenge/2DMOT2015 mot_benchmark\n\n')
|
282 |
+
exit()
|
283 |
+
plt.ion()
|
284 |
+
fig = plt.figure()
|
285 |
+
ax1 = fig.add_subplot(111, aspect='equal')
|
286 |
+
|
287 |
+
if not os.path.exists('output'):
|
288 |
+
os.makedirs('output')
|
289 |
+
pattern = os.path.join(args.seq_path, phase, '*', 'det', 'det.txt')
|
290 |
+
for seq_dets_fn in glob.glob(pattern):
|
291 |
+
mot_tracker = Sort(max_age=args.max_age,
|
292 |
+
min_hits=args.min_hits,
|
293 |
+
iou_threshold=args.iou_threshold) #create instance of the SORT tracker
|
294 |
+
seq_dets = np.loadtxt(seq_dets_fn, delimiter=',')
|
295 |
+
seq = seq_dets_fn[pattern.find('*'):].split(os.path.sep)[0]
|
296 |
+
|
297 |
+
with open(os.path.join('output', '%s.txt'%(seq)),'w') as out_file:
|
298 |
+
print("Processing %s."%(seq))
|
299 |
+
for frame in range(int(seq_dets[:,0].max())):
|
300 |
+
frame += 1 #detection and frame numbers begin at 1
|
301 |
+
dets = seq_dets[seq_dets[:, 0]==frame, 2:7]
|
302 |
+
dets[:, 2:4] += dets[:, 0:2] #convert to [x1,y1,w,h] to [x1,y1,x2,y2]
|
303 |
+
total_frames += 1
|
304 |
+
|
305 |
+
if(display):
|
306 |
+
fn = os.path.join('mot_benchmark', phase, seq, 'img1', '%06d.jpg'%(frame))
|
307 |
+
im =io.imread(fn)
|
308 |
+
ax1.imshow(im)
|
309 |
+
plt.title(seq + ' Tracked Targets')
|
310 |
+
|
311 |
+
start_time = time.time()
|
312 |
+
trackers = mot_tracker.update(dets)
|
313 |
+
cycle_time = time.time() - start_time
|
314 |
+
total_time += cycle_time
|
315 |
+
|
316 |
+
for d in trackers:
|
317 |
+
print('%d,%d,%.2f,%.2f,%.2f,%.2f,1,-1,-1,-1'%(frame,d[4],d[0],d[1],d[2]-d[0],d[3]-d[1]),file=out_file)
|
318 |
+
if(display):
|
319 |
+
d = d.astype(np.int32)
|
320 |
+
ax1.add_patch(patches.Rectangle((d[0],d[1]),d[2]-d[0],d[3]-d[1],fill=False,lw=3,ec=colours[d[4]%32,:]))
|
321 |
+
|
322 |
+
if(display):
|
323 |
+
fig.canvas.flush_events()
|
324 |
+
plt.draw()
|
325 |
+
ax1.cla()
|
326 |
+
|
327 |
+
print("Total Tracking took: %.3f seconds for %d frames or %.1f FPS" % (total_time, total_frames, total_frames / total_time))
|
328 |
+
|
329 |
+
if(display):
|
330 |
+
print("Note: to get real runtime results run without the option: --display")
|
yolov8n.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:31e20dde3def09e2cf938c7be6fe23d9150bbbe503982af13345706515f2ef95
|
3 |
+
size 6534387
|