网站做定向的作用,自助网站建设用什么好,免费网络推广方式,个人网站毕业设计1 内置的代码
在darknet中已经内置了两个py文件 darknet_video.py与darknet_images.py用法类似#xff0c;都是改一改给的参数就行了#xff0c;我们说一下几个关键的参数 input 要预测哪张图像weights 要使用哪个权重config_file 要使用哪个cfg文件data_file 要使用哪个da…1 内置的代码
在darknet中已经内置了两个py文件 darknet_video.py与darknet_images.py用法类似都是改一改给的参数就行了我们说一下几个关键的参数 input 要预测哪张图像weights 要使用哪个权重config_file 要使用哪个cfg文件data_file 要使用哪个data文件thresh 置信度给多少
改完上面这些参数就可以直接运行了 用鼠标点一下图片然后按q可以关闭图片
调用视频有时可以正常用有时不可以会报下面的错可能是opencv的版本问题 这两个代码都很好想的很周道但是看起来太麻烦我通常用下面两个简化版的
2 opencv使用模型
优点只要有python有opencv-python有numpy就能用不需要编译darknet只要有names、cfg、weights就行
缺点只能搞CPU版用摄像头搞效果较差
2.1 识别图像
你需要一张预测图像训练时的classes.names yolov4-tiny-cfg yolov4-tiny_final.weights yolo_opencv.py的内容如下
import cv2
import numpy as npLABELS open(classes.names).read().strip().split(\n)
net cv2.dnn.readNetFromDarknet(yolov4-tiny.cfg, yolov4-tiny_final.weights)
layer net.getUnconnectedOutLayersNames()frame cv2.imread(000003.jpg)
(H, W) frame.shape[:2]
blob cv2.dnn.blobFromImage(frame, 1 / 255.0, (416, 416),swapRBTrue, cropFalse)
net.setInput(blob)
layerOutputs net.forward(layer)
boxes []
confidences []
classIDs []for output in layerOutputs:for detection in output:scores detection[5:]classID np.argmax(scores)confidence scores[classID]box detection[0:4] * np.array([W, H, W, H])(centerX, centerY, width, height) box.astype(int)x int(centerX - (width / 2))y int(centerY - (height / 2))boxes.append([x, y, int(width), int(height)])confidences.append(float(confidence))classIDs.append(classID)idxs cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.3)
if len(idxs) 0:for i in idxs.flatten():(x, y) (boxes[i][0], boxes[i][1])(w, h) (boxes[i][2], boxes[i][3])cv2.rectangle(frame, (x, y), (x w, y h), (0,255,0), 1, lineTypecv2.LINE_AA)text {}: {:.4f}.format(LABELS[classIDs[i]], confidences[i])cv2.putText(frame, text, (x, y - 5), cv2.FONT_HERSHEY_SIMPLEX,0.5, (255,0,0), 1, lineTypecv2.LINE_AA)cv2.imshow(frame,frame)
cv2.waitKey(0)
cv2.destroyAllWindows()
运行后可以显示预测结果 2.2 flask起服务
我们可以搞一个接口来处理识别的功能
服务端
import numpy as np
import cv2
from flask import Flask,request
import base64LABELS open(classes.names).read().strip().split(\n)
net cv2.dnn.readNetFromDarknet(yolov4-tiny.cfg, yolov4-tiny_final.weights)
layer net.getUnconnectedOutLayersNames()app Flask(__name__)
app.route(/predict,methods[POST])
def predict():if request.method POST:image_base64 request.json[img_b64]img base64.b64decode(image_base64)img np.fromstring(img,np.uint8)frame cv2.imdecode(img,cv2.IMREAD_COLOR)(H, W) frame.shape[:2]blob cv2.dnn.blobFromImage(frame, 1 / 255.0, (416, 416),swapRBTrue, cropFalse)net.setInput(blob)layerOutputs net.forward(layer)boxes []confidences []classIDs []for output in layerOutputs:for detection in output:scores detection[5:]classID np.argmax(scores)confidence scores[classID]box detection[0:4] * np.array([W, H, W, H])(centerX, centerY, width, height) box.astype(int)x int(centerX - (width / 2))y int(centerY - (height / 2))boxes.append([x, y, int(width), int(height)])confidences.append(float(confidence))classIDs.append(classID)idxs cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.3)result []if len(idxs) 0:for i in idxs.flatten():(x, y) (boxes[i][0], boxes[i][1])(w, h) (boxes[i][2], boxes[i][3])result_obj {}result_obj[x] xresult_obj[y] yresult_obj[w] wresult_obj[h] hresult_obj[name] LABELS[classIDs[i]]result_obj[confidence] confidences[i]result.append(result_obj)return {result:result}if __name__ __main__:app.run(host192.168.0.105)
客户端
import cv2
import requests
import base64def get_result(url,frame):retval, buffer cv2.imencode(.jpg, frame)image str(base64.b64encode(buffer), utf-8)json_data {img_b64: image}response eval(requests.post(url, jsonjson_data).text).get(result)return response
if __name__ __main__:url http://192.168.0.105:5000/predictframe cv2.imread(000003.jpg)response get_result(url,frame)for result in response:confidence result.get(confidence)x int(result.get(x))y int(result.get(y))w int(result.get(w))h int(result.get(h))name result.get(name)frame cv2.rectangle(frame, (x, y), (x w, y h), (255, 0, 0), 2)frame cv2.putText(frame, name, (x, y), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)cv2.imshow(img,frame)cv2.waitKey(0)cv2.destroyAllWindows()
请求结果 2.3 摄像头识别
原理是图像识别多线程
import cv2
import numpy as np
import queue
import threading
import timeLABELS open(classes.names).read().strip().split(\n)
net cv2.dnn.readNetFromDarknet(yolov4-tiny.cfg, yolov4-tiny_final.weights)
layer net.getUnconnectedOutLayersNames()cap cv2.VideoCapture(0)frame_queue queue.Queue()
detection_result_queue queue.Queue(maxsize1)def video_capture():while cap.isOpened():ret, frame cap.read()if ret:frame_queue.put(frame)cap.release()def predict():while cap.isOpened():start_time time.time()predict_frame frame_queue.get()(H, W) predict_frame.shape[:2]blob cv2.dnn.blobFromImage(predict_frame, 1 / 255.0, (416, 416), swapRBTrue, cropFalse)net.setInput(blob)layerOutputs net.forward(layer)boxes []confidences []classIDs []for output in layerOutputs:for detection in output:scores detection[5:]classID np.argmax(scores)confidence scores[classID]box detection[0:4] * np.array([W, H, W, H])(centerX, centerY, width, height) box.astype(int)x int(centerX - (width / 2))y int(centerY - (height / 2))boxes.append([x, y, int(width), int(height)])confidences.append(float(confidence))classIDs.append(classID)idxs cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.3)if len(idxs) 0:result_list []for i in idxs.flatten():result_dic {}(x, y) (boxes[i][0], boxes[i][1])(w, h) (boxes[i][2], boxes[i][3])label LABELS[classIDs[i]]confidence confidences[i]result_dic.__setitem__(x,x)result_dic.__setitem__(y,y)result_dic.__setitem__(w,w)result_dic.__setitem__(h,h)result_dic.__setitem__(label,label)result_dic.__setitem__(confidence,confidence)result_list.append(result_dic)detection_result_queue.put(result_list)print(time.time()-start_time)cap.release()def draw():while cap.isOpened():draw_frame frame_queue.get()try:predict_results detection_result_queue.get(blockFalse)for predict_result in predict_results:x predict_result.get(x)y predict_result.get(y)w predict_result.get(w)h predict_result.get(h)label predict_result.get(label)confidence predict_result.get(confidence)cv2.rectangle(draw_frame, (x, y), (x w, y h), (0,255,0), 1, lineTypecv2.LINE_AA)text {}: {:.4f}.format(label, confidence)cv2.putText(draw_frame, text, (x, y - 5), cv2.FONT_HERSHEY_SIMPLEX,0.5, (255,0,0), 1, lineTypecv2.LINE_AA)except:passcv2.imshow(draw_frame,draw_frame)cv2.waitKey(1)cap.release()threading.Thread(targetvideo_capture).start()
threading.Thread(targetpredict).start()
threading.Thread(targetdraw).start()
由于是CPU运行效果根据CPU的性能会有差异我是用NX上的6核CPU运行yolov4-tiny大约0.5s一张。这个速度如果在摄像头上的连续识别就非常慢了至少要1秒30帧才能有流畅的感觉也就是0.03秒需要预测一张
0.5/0.03 16.6 或许搞个17线程能够达成流畅的效果没尝试过
3 darknet使用模型
优点可以使用GPU
缺点需要编译darknet才能使用
下面在代码中import darknet指的是import下面这个文件。下面这个py文件依赖了darknet文件夹下的其他文件所以建议把下面的代码放在darknet的根目录下使用 3.1 识别图像
在模型读取的时候是通过data文件找names文件而不是直接找names文件
import cv2
import darknet
import timenetwork, class_names, class_colors darknet.load_network(/home/suyu/darknet/yolo_opencv/yolov7-tiny.cfg,/home/suyu/darknet/custom_training/custom_training.data,/home/suyu/darknet/yolo_opencv/yolov7-tiny_final_origin.weights,batch_size1
)width darknet.network_width(network)
height darknet.network_height(network)
darknet_image darknet.make_image(width, height, 3)image cv2.imread(/home/suyu/darknet/yolo_opencv/280.jpg)
image_rgb cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image_resized cv2.resize(image_rgb, (width, height),interpolationcv2.INTER_LINEAR)
darknet.copy_image_from_bytes(darknet_image, image_resized.tobytes())
start_time time.time()
detections darknet.detect_image(network, class_names, darknet_image, thresh0.8)
print(time.time()-start_time)darknet.free_image(darknet_image)
image darknet.draw_boxes(detections, image_resized, class_colors)
image cv2.cvtColor(image, cv2.COLOR_BGR2RGB)cv2.imshow(image,image)
cv2.waitKey(0)
cv2.destroyAllWindows()
在终端上可以看到对模型进行了读取 画框的颜色每次都是随机的 在预测中如果出现了下面这种情况我们就需要减少预测的结果可以通过标签后面的置信度减少但是我们可以看到有两个框的置信度都在0.99以上这个时候我们就需要用到其他减少框的办法 darknet.detect_image()这个方法除了通过置信度减少预测结果还可以通过非最大值抑制消除冗余也就是最后一个的nms
hier_thresh也是消除冗余的一个参数叫控制层次性阈值(与nms类似)没用过如果nms不行的化再尝试使用它 nms数值越小消除冗余的效果就越好我这里直接改成了0.05 改完之后可以得到还不错的效果 3.2 摄像头识别
原理是识别图像多线程我用yolov7-tiny大概能到0.03秒预测一张这个预测速度不加多线程应该也可以
import cv2
import numpy as np
import queue
import threading
import time
import darknetnetwork, class_names, class_colors darknet.load_network(/home/suyu/darknet/yolo_opencv/yolov7-tiny.cfg,/home/suyu/darknet/custom_training/custom_training.data,/home/suyu/darknet/yolo_opencv/yolov7-tiny_final_origin.weights,batch_size1
)width darknet.network_width(network)
height darknet.network_height(network)
darknet_image darknet.make_image(width, height, 3)cap cv2.VideoCapture(0)frame_queue queue.Queue()
detection_result_queue queue.Queue(maxsize1)def video_capture():while cap.isOpened():ret, frame cap.read()if ret:image_rgb cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)image_resized cv2.resize(image_rgb, (width, height), interpolationcv2.INTER_LINEAR)frame_queue.put(image_resized)cap.release()def predict():while cap.isOpened():start_time time.time()image_resized frame_queue.get()darknet.copy_image_from_bytes(darknet_image, image_resized.tobytes())detections darknet.detect_image(network, class_names, darknet_image, thresh0.8)# darknet.free_image(darknet_image)print(time.time()-start_time)detection_result_queue.put(detections)cap.release()def draw():while cap.isOpened():draw_frame frame_queue.get()# print(draw_frame)try:detections detection_result_queue.get(blockFalse)draw_frame darknet.draw_boxes(detections, draw_frame, class_colors)except:passdraw_frame cv2.cvtColor(draw_frame, cv2.COLOR_BGR2RGB)cv2.imshow(draw_frame,draw_frame)cv2.waitKey(1)cap.release()threading.Thread(targetvideo_capture).start()
threading.Thread(targetpredict).start()
threading.Thread(targetdraw).start()
如果你想对预测的结果进行别的操作你可以用到detection_result_queue中的detections打印出来是这样的 列表套元组元组的第一个值是label第二个值是置信
操作的时候建议多开一个线程然后将detection_result_queue置为2或者更高避免两个线程抢数据的情况
4 onnx使用模型
onnx(Open Neural Network eXchange) 开放式神经网络交换好多模型都可以转换为onnx类型的模型相当于是人工智能模型界的docker了。以高适配性而著名。在性能上并没有优于其他模型。
4.1 用到的库
onnx这个是将其他类型的模型转换为onnx类型的库在arm端上安装可能会有些麻烦但在amd上直接用pip就可以安装了 onnxruntime是跑onnx模型用的onnxruntime在arm端可以直接用pip安装 4.2 darknet的weight转onnx
参考 https://zhuanlan.zhihu.com/p/543345367
源码 GitHub - Tianxiaomo/pytorch-YOLOv4: PyTorch ,ONNX and TensorRT implementation of YOLOv4
安装完onnx与onnxruntime后运行demo_darknet2onnx.py第一个参数是cfg第二个参数是names第三个参数是weights第四个参数是图第五个参数是batch_size直接写1就行了
python demo_darknet2onnx.py /home/suyu/darknet/cfg/yolov4.cfg /home/suyu/darknet/data/coco.names /home/suyu/darknet/yolov4.weights /home/suyu/darknet/data/dog.jpg 1
如果opencv版本过高会爆出下面两个问题是cv2.rectangele()与cv2.putText()的参数需为int 我们需要更改 /pytorch-YOLOv4-master/tool/utils.py
把画红线的地方改成int 成功使用会显示下面这些东西 在文件夹中可以找到通过onnx预测成功的图像和onnx模型 4.3 识别图像
识别的代码是用的上面提供的源码中的东西有的代码直接从utils中复制过来了用的时候不需要引入其他py文件了只需要.onnx文件与.names文件
import sys
import onnx
import os
import argparse
import numpy as np
import cv2
import onnxruntime
import time
import mathdef load_class_names(namesfile):class_names []with open(namesfile, r) as fp:lines fp.readlines()for line in lines:line line.rstrip()class_names.append(line)return class_namesdef nms_cpu(boxes, confs, nms_thresh0.5, min_modeFalse):# print(boxes.shape)x1 boxes[:, 0]y1 boxes[:, 1]x2 boxes[:, 2]y2 boxes[:, 3]areas (x2 - x1) * (y2 - y1)order confs.argsort()[::-1]keep []while order.size 0:idx_self order[0]idx_other order[1:]keep.append(idx_self)xx1 np.maximum(x1[idx_self], x1[idx_other])yy1 np.maximum(y1[idx_self], y1[idx_other])xx2 np.minimum(x2[idx_self], x2[idx_other])yy2 np.minimum(y2[idx_self], y2[idx_other])w np.maximum(0.0, xx2 - xx1)h np.maximum(0.0, yy2 - yy1)inter w * hif min_mode:over inter / np.minimum(areas[order[0]], areas[order[1:]])else:over inter / (areas[order[0]] areas[order[1:]] - inter)inds np.where(over nms_thresh)[0]order order[inds 1]return np.array(keep)def post_processing(img, conf_thresh, nms_thresh, output):# anchors [12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401]# num_anchors 9# anchor_masks [[0, 1, 2], [3, 4, 5], [6, 7, 8]]# strides [8, 16, 32]# anchor_step len(anchors) // num_anchors# [batch, num, 1, 4]box_array output[0]# [batch, num, num_classes]confs output[1]t1 time.time()if type(box_array).__name__ ! ndarray:box_array box_array.cpu().detach().numpy()confs confs.cpu().detach().numpy()num_classes confs.shape[2]# [batch, num, 4]box_array box_array[:, :, 0]# [batch, num, num_classes] -- [batch, num]max_conf np.max(confs, axis2)max_id np.argmax(confs, axis2)t2 time.time()bboxes_batch []for i in range(box_array.shape[0]):argwhere max_conf[i] conf_threshl_box_array box_array[i, argwhere, :]l_max_conf max_conf[i, argwhere]l_max_id max_id[i, argwhere]bboxes []# nms for each classfor j in range(num_classes):cls_argwhere l_max_id jll_box_array l_box_array[cls_argwhere, :]ll_max_conf l_max_conf[cls_argwhere]ll_max_id l_max_id[cls_argwhere]keep nms_cpu(ll_box_array, ll_max_conf, nms_thresh)if (keep.size 0):ll_box_array ll_box_array[keep, :]ll_max_conf ll_max_conf[keep]ll_max_id ll_max_id[keep]for k in range(ll_box_array.shape[0]):bboxes.append([ll_box_array[k, 0], ll_box_array[k, 1], ll_box_array[k, 2], ll_box_array[k, 3], ll_max_conf[k], ll_max_conf[k], ll_max_id[k]])bboxes_batch.append(bboxes)t3 time.time()print(-----------------------------------)print( max and argmax : %f % (t2 - t1))print( nms : %f % (t3 - t2))print(Post processing total : %f % (t3 - t1))print(-----------------------------------)return bboxes_batchdef plot_boxes_cv2(img, boxes, savenameNone, class_namesNone, colorNone):import cv2img np.copy(img)colors np.array([[1, 0, 1], [0, 0, 1], [0, 1, 1], [0, 1, 0], [1, 1, 0], [1, 0, 0]], dtypenp.float32)def get_color(c, x, max_val):ratio float(x) / max_val * 5i int(math.floor(ratio))j int(math.ceil(ratio))ratio ratio - ir (1 - ratio) * colors[i][c] ratio * colors[j][c]return int(r * 255)width img.shape[1]height img.shape[0]for i in range(len(boxes)):box boxes[i]x1 int(box[0] * width)y1 int(box[1] * height)x2 int(box[2] * width)y2 int(box[3] * height)bbox_thick int(0.6 * (height width) / 600)if color:rgb colorelse:rgb (255, 0, 0)if len(box) 7 and class_names:cls_conf box[5]cls_id box[6]print(%s: %f % (class_names[cls_id], cls_conf))classes len(class_names)offset cls_id * 123457 % classesred get_color(2, offset, classes)green get_color(1, offset, classes)blue get_color(0, offset, classes)if color is None:rgb (red, green, blue)msg str(class_names[cls_id]) str(round(cls_conf,3))t_size cv2.getTextSize(msg, 0, 0.7, thicknessbbox_thick // 2)[0]c1, c2 (x1,y1), (x2, y2)c3 (c1[0] t_size[0], c1[1] - t_size[1] - 3)cv2.rectangle(img, (int(x1),int(y1)), (int(np.float32(c3[0])), int(np.float32(c3[1]))), rgb, -1)img cv2.putText(img, msg, (int(c1[0]), int(np.float32(c1[1] - 2))), cv2.FONT_HERSHEY_SIMPLEX,0.7, (0,0,0), bbox_thick//2,lineTypecv2.LINE_AA)img cv2.rectangle(img, (int(x1), int(y1)), (int(x2), int(y2)), rgb, bbox_thick)if savename:print(save plot results to %s % savename)cv2.imwrite(savename, img)return imgdef detect(session, image_src, namesfile):IN_IMAGE_H session.get_inputs()[0].shape[2]IN_IMAGE_W session.get_inputs()[0].shape[3]# Inputresized cv2.resize(image_src, (IN_IMAGE_W, IN_IMAGE_H), interpolationcv2.INTER_LINEAR)img_in cv2.cvtColor(resized, cv2.COLOR_BGR2RGB)img_in np.transpose(img_in, (2, 0, 1)).astype(np.float32)img_in np.expand_dims(img_in, axis0)img_in / 255.0#print(Shape of the network input: , img_in.shape)# Computeinput_name session.get_inputs()[0].nameoutputs session.run(None, {input_name: img_in})boxes post_processing(img_in, 0.4, 0.6, outputs)class_names load_class_names(namesfile)return plot_boxes_cv2(image_src, boxes[0], class_namesclass_names)if __name__ __main__:session onnxruntime.InferenceSession(yolov4_1_3_640_640_static.onnx)namesfile classes.namesimage_src cv2.imread(3.png)detected_img detect(session, image_src, namesfile)cv2.imshow(detected_img,detected_img)cv2.waitKey(0)cv2.destroyAllWindows() 4.4 摄像头识别
用的采集、预测、画 三线程。模型输入width与height都为320两分类在树莓派4B中大概是0.5s一张大概是1秒2个识别帧
import cv2
import numpy as np
import queue
import threading
import timeimport sys
import onnx
import os
import argparse
import onnxruntime
import mathdef load_class_names(namesfile):class_names []with open(namesfile, r) as fp:lines fp.readlines()for line in lines:line line.rstrip()class_names.append(line)return class_namesdef nms_cpu(boxes, confs, nms_thresh0.5, min_modeFalse):x1 boxes[:, 0]y1 boxes[:, 1]x2 boxes[:, 2]y2 boxes[:, 3]areas (x2 - x1) * (y2 - y1)order confs.argsort()[::-1]keep []while order.size 0:idx_self order[0]idx_other order[1:]keep.append(idx_self)xx1 np.maximum(x1[idx_self], x1[idx_other])yy1 np.maximum(y1[idx_self], y1[idx_other])xx2 np.minimum(x2[idx_self], x2[idx_other])yy2 np.minimum(y2[idx_self], y2[idx_other])w np.maximum(0.0, xx2 - xx1)h np.maximum(0.0, yy2 - yy1)inter w * hif min_mode:over inter / np.minimum(areas[order[0]], areas[order[1:]])else:over inter / (areas[order[0]] areas[order[1:]] - inter)inds np.where(over nms_thresh)[0]order order[inds 1]return np.array(keep)def post_processing(img, conf_thresh, nms_thresh, output):box_array output[0]confs output[1]if type(box_array).__name__ ! ndarray:box_array box_array.cpu().detach().numpy()confs confs.cpu().detach().numpy()num_classes confs.shape[2]box_array box_array[:, :, 0]max_conf np.max(confs, axis2)max_id np.argmax(confs, axis2)bboxes_batch []for i in range(box_array.shape[0]):argwhere max_conf[i] conf_threshl_box_array box_array[i, argwhere, :]l_max_conf max_conf[i, argwhere]l_max_id max_id[i, argwhere]bboxes []for j in range(num_classes):cls_argwhere l_max_id jll_box_array l_box_array[cls_argwhere, :]ll_max_conf l_max_conf[cls_argwhere]ll_max_id l_max_id[cls_argwhere]keep nms_cpu(ll_box_array, ll_max_conf, nms_thresh)if (keep.size 0):ll_box_array ll_box_array[keep, :]ll_max_conf ll_max_conf[keep]ll_max_id ll_max_id[keep]for k in range(ll_box_array.shape[0]):bboxes.append([ll_box_array[k, 0], ll_box_array[k, 1], ll_box_array[k, 2], ll_box_array[k, 3], ll_max_conf[k], ll_max_conf[k], ll_max_id[k]])bboxes_batch.append(bboxes)return bboxes_batchdef get_color(c, x, max_val):colors np.array([[1, 0, 1], [0, 0, 1], [0, 1, 1], [0, 1, 0], [1, 1, 0], [1, 0, 0]], dtypenp.float32)ratio float(x) / max_val * 5i int(math.floor(ratio))j int(math.ceil(ratio))ratio ratio - ir (1 - ratio) * colors[i][c] ratio * colors[j][c]return int(r * 255)def detect(session, image_src, namesfile):IN_IMAGE_H session.get_inputs()[0].shape[2]IN_IMAGE_W session.get_inputs()[0].shape[3]# Inputresized cv2.resize(image_src, (IN_IMAGE_W, IN_IMAGE_H), interpolationcv2.INTER_LINEAR)img_in cv2.cvtColor(resized, cv2.COLOR_BGR2RGB)img_in np.transpose(img_in, (2, 0, 1)).astype(np.float32)img_in np.expand_dims(img_in, axis0)img_in / 255.0# Computeinput_name session.get_inputs()[0].nameoutputs session.run(None, {input_name: img_in})boxes post_processing(img_in, 0.4, 0.6, outputs)return boxes[0]session onnxruntime.InferenceSession(yolov4_1_3_640_640_static.onnx)
namesfile classes.names
class_names load_class_names(namesfile)cap cv2.VideoCapture(0)frame_queue queue.Queue()
detection_result_queue queue.Queue(maxsize1)def video_capture():while cap.isOpened():ret, frame cap.read()if ret:frame_queue.put(frame)cap.release()def predict():while cap.isOpened():start_time time.time()predict_frame frame_queue.get()result_list detect(session, predict_frame, namesfile)detection_result_queue.put(result_list)print(time.time()-start_time)cap.release()def draw():while cap.isOpened():draw_frame frame_queue.get()try:boxes detection_result_queue.get(blockFalse)img draw_framewidth img.shape[1]height img.shape[0]for i in range(len(boxes)):box boxes[i]x1 int(box[0] * width)y1 int(box[1] * height)x2 int(box[2] * width)y2 int(box[3] * height)bbox_thick int(0.6 * (height width) / 600)rgb (255, 0, 0)if len(box) 7 and class_names:cls_conf box[5]cls_id box[6]print(%s: %f % (class_names[cls_id], cls_conf))classes len(class_names)offset cls_id * 123457 % classesred get_color(2, offset, classes)green get_color(1, offset, classes)blue get_color(0, offset, classes)rgb (red, green, blue)msg str(class_names[cls_id]) str(round(cls_conf,3))t_size cv2.getTextSize(msg, 0, 0.7, thicknessbbox_thick // 2)[0]c1, c2 (x1,y1), (x2, y2)c3 (c1[0] t_size[0], c1[1] - t_size[1] - 3)cv2.rectangle(img, (int(x1),int(y1)), (int(np.float32(c3[0])), int(np.float32(c3[1]))), rgb, -1)img cv2.putText(img, msg, (int(c1[0]), int(np.float32(c1[1] - 2))), cv2.FONT_HERSHEY_SIMPLEX,0.7, (0,0,0), bbox_thick//2,lineTypecv2.LINE_AA)img cv2.rectangle(img, (int(x1), int(y1)), (int(x2), int(y2)), rgb, bbox_thick)draw_frame img#except Exception as e:#print(e)except:passcv2.imshow(draw_frame,draw_frame)cv2.waitKey(1)cap.release()threading.Thread(targetvideo_capture).start()
threading.Thread(targetpredict).start()
threading.Thread(targetdraw).start()
在树莓派4B上是下面这个效果
