replace sort to deepsort

app.py

@@ -9,6 +9,7 @@ import tempfile #创建输出临时文件夹
 import os
 from detectMotion import * #单独的运动检测
 import time
+import deep_sort.deep_sort.deep_sort as ds
 
 
 #导入YoloV8模型,模型名称存储在model_list.txt中
@@ -33,8 +34,6 @@ classNames=['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train'
     'scissors',
     'teddy bear', 'hair drier', 'toothbrush']
 
-#运动检测算法参数
-tracker=Sort(max_age=20,min_hits=3,iou_threshold=0.3)
 
 #实时显示人流量统计图表
 def plot_number():
@@ -44,11 +43,6 @@ def plot_number():
     plt.xlabel('Second')
     plt.ylabel('Traffic')
     plt.grid(linestyle='--', alpha=0.3, linewidth=2)
-    #若函数在YOLO模型生成bounding box前运行，则返回空图表
-    if number_of_people_in_one_frame_list == []:
-        time.sleep(3) #延迟3秒
-        plt.plot([])
-        return fig
     plt.plot(number_of_people_in_one_frame_list)
     return fig
 
@@ -100,6 +94,7 @@ def processImg(img,sigma,median_filter,ISR):
 #视频处理
 def processVideo(inputPath, codec, model):
     global number_of_people_in_one_frame_list
+    tracker = ds.DeepSort('deep_sort/deep_sort/deep/checkpoint/ckpt.t7')
     if inputPath == None:
         raise gr.Error("请先上传视频")
     model=YOLO(model)
@@ -138,7 +133,8 @@ def processVideo(inputPath, codec, model):
     while True:
         ret, img = cap.read()#将每一帧图片读取到img当中
         results=model(img,stream=True)#使用YoloV8模型进行推理
-        detections=np.empty((0, 5))#初始化运动检测
+        detections=np.empty((0, 4))#初始化运动检测
+        confarray = []
         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
             break
         img, __, __= hisEqulColor(img)#视频增强
@@ -147,52 +143,40 @@ def processVideo(inputPath, codec, model):
             boxes=r.boxes
             for box in boxes:
                 #读取每一帧识别出的边界信息，并显示
-                x1,y1,x2,y2=box.xyxy[0]
+                x1,y1,x2,y2=box.xywh[0]
                 x1,y1,x2,y2=int(x1),int(y1),int(x2),int(y2)#将tensor类型转变为整型
                 conf=math.ceil(box.conf[0]*100)/100#对conf取2位小数
                 cls=int(box.cls[0])#获取物体类别标签
-                #当标签为人，且可信度大于0.3的时候，将人标识出来
-                if cls==0 and conf > 0.3:
+                #只检测人
+                if cls==0:
                     number_of_people_in_one_frame += 1
-                    cv2.rectangle(img,(x1,y1),(x2,y2),(255,0,255),3)
-                    #print(conf)
-                    #cvzone.putTextRect(img,f'{classNames[cls]}{conf}',(max(0,x1),max(30,y1)),scale=0.7,thickness=1)
-                    currentArray=np.array([x1,y1,x2,y2,conf])
+                    currentArray=np.array([x1,y1,x2,y2])
+                    confarray.append(conf)
                     detections=np.vstack((detections,currentArray))#按行堆叠数据
         sum_of_frame += 1
         if sum_of_frame % fps == 0:
             number_of_people_in_one_frame_list.append(number_of_people_in_one_frame)
         number_of_people_in_one_frame = 0
         #运动检测
-        resultsTracker=tracker.update(detections)
+        resultsTracker=tracker.update(detections, confarray, img)
         for result in resultsTracker:
             x1,y1,x2,y2,Id=result
             number_of_people=max(str(int(Id)),str(number_of_people))
             x1,y1,x2,y2=int(x1),int(y1),int(x2),int(y2)#将浮点数转变为整型
-            cvzone.putTextRect(img,f'{int(Id)}',(max(0,x1),max(30,y1)),scale=0.7,thickness=1)
+            cv2.rectangle(img,(x1,y1),(x2,y2),(255,0,255),3)
+            cvzone.putTextRect(img,f'{int(Id)}',(max(-10,x1),max(40,y1)),scale=1.3,thickness=2)
         #image_np = np.squeeze(img.render())#用np.squeeze将输出结果降维
         output_viedo.write(img)#将处理后的图像写入视频
     output_viedo.release()#释放
     cap.release()#释放
-    return video_save_path,video_save_path
+    return video_save_path,video_save_path,number_of_people
 
 
-
-#WebUi图形界面（interface）
-#demo = gr.Interface(
-#    fn=processVideo,
-#    inputs=["text","text"],
-#    outputs=["text"],
-#    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"]],
-#    title="运动检测与行人跟踪",
-#    description="请输入绝对路径"
-#)
-
 #WebUi图形界面（block）
 with gr.Blocks() as demo:
     gr.Markdown("""
     # 运动检测与行人跟踪
-    基于opencv + yoloV8 + sort
+    基于opencv + yoloV8 + deepsort
     """)
     with gr.Tab("视频识别"):
         with gr.Row():
@@ -205,6 +189,7 @@ with gr.Blocks() as demo:
             with gr.Column():
                 text_output = gr.Video()
                 text_output_path = gr.Text(label="输出路径")
+                total_ID = gr.Text(label="总人数")
         with gr.Row():
             with gr.Column():
                 figure_number_output = gr.Plot(label="人流量")
@@ -237,13 +222,13 @@ with gr.Blocks() as demo:
     
 
     videoProcess_button.click(processVideo, inputs=[text_inputPath, codec, model],
-                               outputs=[text_output,text_output_path])
+                               outputs=[text_output,text_output_path, total_ID])
     image_button.click(processImg, inputs=[image_input,image_sigma,median_filter,Add_ISR], 
                        outputs=[image_output,figure_pre_output,figure_post_output])
     motionProcess_button.click(motionDetection, inputs=[motion_inputPath], 
                                outputs=[motion_output_frame,motion_output_fmask,
                                         frame_output_path,fmask_output_path])
-    videoProcess_button.click(plot_number,outputs=figure_number_output,every=1)
+    videoProcess_button.click(plot_number,outputs=figure_number_output,every=2)
 
 demo.queue()#当有多个请求时，排队
 demo.launch()#生成内网链接，如需要公网链接，括号内输入share=True