OpenCV計(jì)算機(jī)視覺(jué)實(shí)戰(zhàn)停車場(chǎng)車位識(shí)別?。ㄍ暾a）

1、OpenCV 計(jì)算機(jī)視覺(jué)實(shí)戰(zhàn)，停車場(chǎng)車位識(shí)別?。ㄍ暾a）！src1:圖1 src2:圖2 mask：圖1和圖2與操作的掩碼輸出圖像def select_rgb_white_yellow(self,image):masked = cv2.bitwise_and(image, image, mask = white_mask)self.cv_show(masked,masked)def convert_gray_scale(self,image):cv2.circle(img,中點(diǎn)(x,y),半徑r,color,粗細(xì)) 根據(jù)給定的圓和半徑等畫圓 畫出指定點(diǎn)def filter_region(se

2、lf,image, vertices):剔除掉不需要的地vertices = np.array(pt_1, pt_2, pt_3, pt_4, pt_5, pt_6, dtype=32)point_img = image.copy()point_img = cv2.cvtColor(point_img, cv2.COLOR_GRAY2RGB)for point in vertices0:return self.filter_region(image, vertices)abs(y2-y1) =25 and abs(x2-x1) = 55 長(zhǎng)度太長(zhǎng)的也不要def hough_line

3、s(self,image):# 輸?shù)膱D像需要是邊緣檢測(cè)后的結(jié)果if make_copy:image = np.copy(image)cleaned = for line in lines:for x1,y1,x2,y2 in line:step 3:指定間距于10的,劃分為不同的列,共12簇def identify_blocks(self,image, lines, make_copy=True):if make_copy:if abs(y2-y1) =25 and abs(x2-x1) = 55:cleaned.append(x1,y1,x2,y2)clusters = dIndex = 0

4、distance = abs(list1i+10 - list1i0)if distance = clus_dist:if not dIndex in clusters.keys(): clustersdIndex = clustersdIndex.append(list1i)clustersdIndex.append(list1i + 1)else:dIndex += 1#Step 4: 得到坐標(biāo)rects = i = 0for key in clusters:avg_x2 = 0for tup in cleaned:avg_x1 += tup0avg_x2 += tup2buff = 7f

5、or key in rects:tup_topLeft = (int(rectskey0 - buff), int(rectskey1)tup_botRight = (int(rectskey2 + buff), int(rectskey3)cv2.rectangle(new_image, tup_topLeft,tup_botRight,(0,255,0),3)return new_image, rectsspot_dict = # 字典：個(gè)車位對(duì)應(yīng)個(gè)位置adj_x1 = 0: -8, 1:-15, 2:-15, 3:-15, 4:-15, 5:-15, 6:-15, 7:-15, 8:-1

6、0, 9:-10, 10:-10, 11:0adj_x2 = 0: 0, 1: 15, 2:15, 3:15, 4:15, 5:15, 6:15, 7:15, 8:10, 9:10, 10:10, 11:0# 繼續(xù)微調(diào)cv2.rectangle(new_image, (x1, y1),(x2,y2),(0,255,0),2)# (y2-y1)/gap表能停多少輛車num_splits = int(abs(y2-y1)/gap)for i in range(0, num_splits+1):y = int(y1 + i*gap)x = int(x1 + x2)/2)if key = 0 or k

7、ey = (len(rects) -1):tot_spots += num_splits +1else:tot_spots += 2*(num_splits +1) # 2# 字典對(duì)應(yīng)好if key = 0 or key = (len(rects) -1):for i in range(0, num_splits+1):cur_len = len(spot_dict)y = int(y1 + i*gap)x = int(x1 + x2)/2)filename = with_parking.jpgcv2.imwrite(filename, new_image)return new_image,

8、spot_dictsave_images_for_cnn 保存所有切割出來(lái)的圖必須的步驟，主要是要獲得車位坐標(biāo)的字典def save_images_for_cnn(self,image, spot_dict, folder_name =cnn_data):for spot in spot_dict.keys():filename = spot + str(spot_id) +.jpgprint(spot_img.shape, filename, (x1,x2,y1,y2)cv2.imwrite(os.path.join(folder_name, filename), spot_img)fina

9、l_spot_dict 是img_process函數(shù) return的車位坐標(biāo)字典if _name_ = _main_:class_dictionary0 = emptyclass_dictionary1 = occupiedpark = Parking() # Parking對(duì)象park.show_images(test_images)final_spot_dict = img_process(test_images,park) # 圖像處理model = keras_model(weights_path)img_test(test_images,final_spot_dict,model,c

10、lass_dictionary)video_test(video_name,final_spot_dict,model,class_dictionary)from keras.applications.imagenet_utils import preprocess_inputfrom keras.models import load_modelfrom keras.preprocessing import imagefrom Parking import Parkingimport picklecwd = os.getcwd()def img_process(test_images,park

11、):white_yellow_images = list(map(park.select_rgb_white_yellow, test_images)park.show_images(white_yellow_images)list_of_lines = list(map(park.hough_lines, roi_images)line_images = rect_images = rect_coords = # 區(qū)域置空spot_pos = for image, rects in zip(test_images, rect_coords):for image, rects in zip(t

12、est_images, rect_coords):new_image, spot_dict = park.draw_parking(image, rects)delineated.append(new_image)pickle.dump(final_spot_dict, handle, protocol=pickle.HIGHEST_PROTOCOL)# park.save_images_for_cnn(test_images0,final_spot_dict)model = load_model(weights_path)def img_test(test_images,final_spot

13、_dict,model,class_dictionary):for i in range (len(test_images):class_dictionary0 = emptyclass_dictionary1 = occupiedpark = Parking() # Parking對(duì)象park.show_images(test_images)final_spot_dict = img_process(test_images,park) # 圖像處理model = keras_model(weights_path)1234567891011121314151617181920212223242

14、52627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384import os, globrows = (len(images)+1)/colsplt.figure(figsize=(15, 12)for i, image in enumerate(images):plt.subplot(rows, cols, i+1)cmap = gray if len(image.shape)=2 else cmapplt.ims

15、how(image, cmap=cmap)plt.xticks()plt.tight_layout(pad=0, h_pad=0, w_pad=0)plt.show()cv2.waitKey(0)def select_rgb_white_yellow(self,image):# 過(guò)濾掉背景l(fā)ower = np.uint8(120, 120, 120)masked = cv2.bitwise_and(image, image, mask = white_mask)self.cv_show(masked,masked)def convert_gray_scale(self,image):def f

16、ilter_region(self,image, vertices):動(dòng)選擇區(qū)域# first, define the polygon by verticesrows, cols = image.shape:2pt_1 = cols*0.05, rows*0.90pt_2 = cols*0.05, rows*0.70pt_3 = cols*0.30, rows*0.55pt_4 = cols*0.6, rows*0.15pt_5 = cols*0.90, rows*0.15pt_6 = cols*0.90, rows*0.90vertices = np.array(pt_1, pt_2, pt

17、_3, pt_4, pt_5, pt_6, dtype=32)point_img = image.copy()return self.filter_region(image, vertices)def hough_lines(self,image):# 輸?shù)膱D像需要是邊緣檢測(cè)后的結(jié)果for line in lines:if abs(y2-y1) =25 and abs(x2-x1) = 55:cleaned.append(x1,y1,x2,y2)cv2.line(image, (x1, y1), (x2, y2), color, thickness)print( No lines

18、detected: , len(cleaned)return imagedef identify_blocks(self,image, lines, make_copy=True):if make_copy:new_image = np.copy(image)#Step 1: 過(guò)濾部分直線if abs(y2-y1) =25 and abs(x2-x1) = 55:cleaned.append(x1,y1,x2,y2)#Step 2: x1進(jìn)排序import operatordistance = abs(list1i+10 - list1i0)if distance = clus_dist:if

19、 not dIndex in clusters.keys(): clustersdIndex = clustersdIndex.append(list1i)clustersdIndex.append(list1i + 1)else:dIndex += 1#Step 4: 得到坐標(biāo)rects = i = 0for key in clusters:avg_x2 = 0for tup in cleaned:avg_x1 += tup0avg_x2 += tup2tup_topLeft = (int(rectskey0 - buff), int(rectskey1)tup_botRight = (in

20、t(rectskey2 + buff), int(rectskey3)cv2.rectangle(new_image, tup_topLeft,tup_botRight,(0,255,0),3)return new_image, rectscv2.rectangle(new_image, (x1, y1),(x2,y2),(0,255,0),2)# (y2-y1)/gap 表能停多少輛車num_splits = int(abs(y2-y1)/gap)for i in range(0, num_splits+1):y = int(y1 + i*gap)x = int(x1 + x2)/2)els

21、e:tot_spots += 2*(num_splits +1) # 2# 字典對(duì)應(yīng)好x = int(x1 + x2)/2)filename = with_parking.jpgcv2.imwrite(filename, new_image)return new_image, spot_dictnew_image = np.copy(image)for spot in spot_dict.keys():def save_images_for_cnn(self,image, spot_dict, folder_name =cnn_data):for spot in spot_dict.keys(

22、):def make_prediction(self,image,model,class_dictionary):image = np.expand_dims(img, axis=0)class_predicted = model.predict(image)new_image = np.copy(image)overlay = np.copy(image)self.cv_show(new_image,new_image)cnt_empty = 0all_spots = 0(x1, y1, x2, y2) = spotlabel = self.make_prediction(spot_img,model,class_dictionary)if label = empty:cv2.addWeighted(overlay, alpha, new_image, 1 - alpha, 0, new_image)cv2.putText(new_image, Available: %d spots %cnt_empty, (30, 95),cv2.FONT_HERSHEY_SIMPLEX,0.7, (255, 255, 255), 2)cv2.putText(new_image, Total: %d spots %all_spots, (30, 12