wangjin0928
pirms 1 gada
4 mainītis faili ar 345 papildinājumiem un 94 dzēšanām
+ 0
- 1
AI.py
Parādīt failu
| @@ -117,7 +117,6 @@ def AI_process(im0s,model,segmodel,names,label_arraylist,rainbows,objectPar={ 'h | |||
| postPar['label_info']=label_info | |||
| tlist = list(postPar.keys()); tlist.sort() | |||
| if len(det_coords)> 0: | |||
| list8, image,time_infos = trafficPostProcessingV2(postPar) | |||
| Accident_results = np.array(list8,dtype=object) | |||
| acc_det=[] | |||
Binārs
__pycache__/AI.cpython-38.pyc
Parādīt failu
Binārs
segutils/__pycache__/trafficUtils.cpython-38.pyc
Parādīt failu
+ 345
- 93
segutils/trafficUtils.py
Parādīt failu
| @@ -651,7 +651,7 @@ def trafficPostProcessingV2_1(traffic_dict): | |||
| t10 = time.time() | |||
| time_infos = 'postTime:%.2f ( predMapBinaryTime:%.2f , findContours:%.2f ruleJudge:%.2f coorsResize:%.2f )' %(get_ms(t10,t3), get_ms(t4,t3),get_ms(t6,t5),get_ms(t8,t7),get_ms(t10,t9) ) | |||
| return list8, list11, image,time_infos | |||
| def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| def trafficPostProcessingV6(traffic_dict): | |||
| """ | |||
| 对于字典traffic_dict中的各个键，说明如下： | |||
| speedRoadArea：speedRoad的最小外接矩形的面积 | |||
| @@ -667,6 +667,245 @@ def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| 最终输出格式：[[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, max([角度得分, 长宽比得分, 最小距离得分]), 交通事故类别], ...] | |||
| 交通事故类别：0表示角度，1表示长宽比，2表示最短距离，3表示未发生交通事故 | |||
| """ | |||
| det_cors=[] | |||
| for bb in traffic_dict['det']: | |||
| det_cors.append( (int(bb[1]), int(bb[2])) ) | |||
| det_cors.append( (int(bb[3]), int(bb[4])) ) | |||
| traffic_dict['vehicleCoordinate'] = det_cors | |||
| t3 = time.time() | |||
| list17 = [] | |||
| list21 = [] # 存储一副图像中vehicles的contours | |||
| image_speedRoad = traffic_dict['mask'].copy() | |||
| image_vehicle = traffic_dict['mask'].copy() | |||
| image_speedRoad[image_speedRoad == 2] = 0 # 将vehicle过滤掉，只包含背景和speedRoad | |||
| image_vehicle[image_vehicle == 1] = 0 # 将speedRoad过滤掉，只包含背景和vehicle | |||
| for key in traffic_dict['label_info']: | |||
| list17.append(traffic_dict['label_info'][key]) | |||
| colour_codes = np.array(list17) # [[0 0 0],[128 0 0],[0 128 0]] ndarray类型 | |||
| preds_squeeze_predict_speedRoad = colour_codes[image_speedRoad] | |||
| preds_squeeze_predict_vehicle = colour_codes[image_vehicle] | |||
| preds_squeeze_predict = colour_codes[traffic_dict['mask']] | |||
| image = cv2.cvtColor(np.uint8(preds_squeeze_predict), cv2.COLOR_RGB2BGR) | |||
| image_speedRoad = cv2.cvtColor(np.uint8(preds_squeeze_predict_speedRoad), cv2.COLOR_RGB2BGR) # 道路 | |||
| image_vehicle = cv2.cvtColor(np.uint8(preds_squeeze_predict_vehicle), cv2.COLOR_RGB2BGR) # 车辆 | |||
| t4 = time.time() | |||
| img1 = cv2.cvtColor(image_speedRoad, cv2.COLOR_BGR2GRAY) | |||
| contours, hierarchy = cv2.findContours(img1, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) | |||
| t5 = time.time() | |||
| list1 = [] # 过渡使用 | |||
| list2 = [] # 存放道路坐标(Xmin,Xmax,Ymin,Ymax) | |||
| list3 = [] # # 存储vehicle最小外接矩形的最短边 | |||
| list4 = [] # list4存储车辆的box参数 | |||
| list7 = [] # 存储vehicle的宽高 | |||
| for cnt in contours: # 道路 | |||
| rect = cv2.minAreaRect(cnt) | |||
| if rect[1][0] * rect[1][1] > traffic_dict['speedRoadArea']: # 过滤掉面积小于阈值的speedRoad | |||
| box = cv2.boxPoints(rect).astype(np.int32) | |||
| Xmin = min(box[0][0], box[1][0], box[2][0], box[3][0]) | |||
| Xmax = max(box[0][0], box[1][0], box[2][0], box[3][0]) | |||
| Ymin = min(box[0][1], box[1][1], box[2][1], box[3][1]) | |||
| Ymax = max(box[0][1], box[1][1], box[2][1], box[3][1]) | |||
| list1.append(Xmin) # 将道路矩形框四个顶点的Xmin,Xmax,Ymin,Ymax存储在list1中 | |||
| list1.append(Xmax) | |||
| list1.append(Ymin) | |||
| list1.append(Ymax) | |||
| list1.append(rect[2]) # 将道路矩形框与水平方向的夹角存储在list1中 | |||
| list1.append(rect[1]) # 将道路的宽高存储在list1中 | |||
| list2.append(list1) | |||
| list1 = [] | |||
| for i in range(0, len(traffic_dict['vehicleCoordinate']), 2): | |||
| mask = np.zeros(image_vehicle.shape[:2], dtype="uint8") | |||
| x0 = int(traffic_dict['vehicleCoordinate'][i][0] * traffic_dict['ZoomFactor']['x']) | |||
| y0 = int(traffic_dict['vehicleCoordinate'][i][1] * traffic_dict['ZoomFactor']['y']) | |||
| x1 = int(traffic_dict['vehicleCoordinate'][i + 1][0] * traffic_dict['ZoomFactor']['x']) | |||
| y1 = int(traffic_dict['vehicleCoordinate'][i + 1][1] * traffic_dict['ZoomFactor']['y']) | |||
| cv2.rectangle(mask, (x0, y0), (x1, y1), 255, -1, lineType=cv2.LINE_AA) | |||
| image_vehicle_masked = cv2.bitwise_and(image_vehicle, image_vehicle, mask=mask) | |||
| img2 = cv2.cvtColor(image_vehicle_masked, cv2.COLOR_BGR2GRAY) | |||
| contours2, hierarchy2 = cv2.findContours(img2, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) | |||
| list21.append(contours2) | |||
| for i in range(len(list21)): | |||
| for cnt in list21[i]: | |||
| flag = False | |||
| rect = cv2.minAreaRect(cnt) | |||
| box = cv2.boxPoints(rect).astype(np.int32) | |||
| if rect[1][0] * rect[1][1] > traffic_dict['vehicleArea']: # 过滤掉面积小于阈值的vehicle | |||
| list3.append(min(rect[1])) | |||
| list4.append(box) | |||
| list7.append(rect[1]) | |||
| for j in range(len(list2)): | |||
| if (rect[0][0] > list2[j][0] and rect[0][0] < list2[j][1]) and (rect[0][1] > list2[j][2] and rect[0][1] < list2[j][3]): # 判断车辆矩形框的中心点坐标是否在道路矩形框Xmin,Xmax,Ymin和Ymax的范围内； | |||
| box = cv2.boxPoints(rect).astype(np.int32) # 将Box2D结构作为输入并返回4个角点。 | |||
| if (box[0][0] >= list2[j][0] and box[0][0] <= list2[j][1] and box[0][1] >= list2[j][2] and box[0][1] <= list2[j][3]) and (box[1][0] >= list2[j][0] and box[1][0] <= list2[j][1] and box[1][1] >= list2[j][2] and box[1][1] <= list2[j][3]) and (box[2][0] >= list2[j][0] and box[2][0] <= list2[j][1] and box[2][1] >= list2[j][2] and box[2][1] <= list2[j][3]) and (box[3][0] >= list2[j][0] and box[3][0] <= list2[j][1] and box[3][1] >= list2[j][2] and box[3][1] <= list2[j][3]): # 比较车辆矩形框四个顶点的坐标是否都在道路矩形框Xmin,Xmax,Ymin和Ymax的范围内，若都在，则说明该车辆在这条道路上。 | |||
| y_min = min(box[0][1], box[1][1], box[2][1], box[3][1]) | |||
| y_max = max(box[0][1], box[1][1], box[2][1], box[3][1]) | |||
| angle = abs(rect[2] - list2[j][4]) | |||
| roundness = min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1]) | |||
| traffic_dict['det'][i].append(angle) | |||
| traffic_dict['det'][i].append(roundness) | |||
| traffic_dict['det'][i].append(999) # 给最小距离占位 | |||
| traffic_dict['det'][i].append([-1, -1, -1]) | |||
| traffic_dict['det'][i].append(666) # 给事故类别占位 | |||
| if abs(rect[2] - list2[j][4]) >= traffic_dict['speedRoadVehicleAngleMin'] and abs( | |||
| rect[2] - list2[j][4]) <= traffic_dict[ | |||
| 'speedRoadVehicleAngleMax']: # 当道路同水平方向的夹角与车辆同水平方向的夹角的差值在15°和75°之间时，需要将车辆框出来 | |||
| score1 = float(abs(rect[2] - list2[j][4]) / 90) | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| if min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1]) > traffic_dict['roundness']: | |||
| score2 = (min(rect[1][0], rect[1][1]) - max(rect[1][0], rect[1][1]) * traffic_dict[ | |||
| 'roundness']) / (max(rect[1][0], rect[1][1]) * ( | |||
| 1 - traffic_dict['roundness'])) | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| elif list2[j][5][0] < list2[j][5][1]: # speedRoad的最小外接矩形的w<h，说明该speedRoad是左侧道路 | |||
| if y_min > 0 and y_max < image.shape[0]: # 过滤掉上下方被speedRoad的边界截断的vehicle | |||
| if abs(rect[2] - list2[j][4]) >= 0 and abs(rect[2] - list2[j][4]) < traffic_dict[ | |||
| 'speedRoadVehicleAngleMin']: | |||
| if (rect[1][0] >= rect[1][1]) or ( | |||
| min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1])) >= \ | |||
| traffic_dict['roundness']: | |||
| if rect[2] == list2[j][4]: | |||
| score1 = 0.10 | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| else: | |||
| score1 = float(abs(rect[2] - list2[j][4]) / 90) | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| if min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1]) > traffic_dict[ | |||
| 'roundness']: | |||
| score2 = (min(rect[1][0], rect[1][1]) - max(rect[1][0], rect[1][1]) * | |||
| traffic_dict['roundness']) / ( | |||
| max(rect[1][0], rect[1][1]) * ( | |||
| 1 - traffic_dict['roundness'])) | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| elif abs(rect[2] - list2[j][4]) > traffic_dict['speedRoadVehicleAngleMax'] and abs( | |||
| rect[2] - list2[j][4]) <= 90: | |||
| if rect[1][0] <= rect[1][1] or ( | |||
| min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1])) >= \ | |||
| traffic_dict['roundness']: | |||
| score1 = float(abs(rect[2] - list2[j][4]) / 90) | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| if min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1]) > traffic_dict[ | |||
| 'roundness']: | |||
| score2 = (min(rect[1][0], rect[1][1]) - max(rect[1][0], rect[1][1]) * | |||
| traffic_dict['roundness']) / ( | |||
| max(rect[1][0], rect[1][1]) * ( | |||
| 1 - traffic_dict['roundness'])) | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| elif list2[j][5][0] > list2[j][5][1]: # speedRoad的最小外接矩形的w>h，说明该speedRoad是右侧道路 | |||
| if y_min > 0 and y_max < image.shape[0]: | |||
| if abs(rect[2] - list2[j][4]) >= 0 and abs(rect[2] - list2[j][4]) < traffic_dict[ | |||
| 'speedRoadVehicleAngleMin']: | |||
| if rect[1][0] <= rect[1][1] or ( | |||
| min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1])) >= \ | |||
| traffic_dict['roundness']: | |||
| if rect[2] == list2[j][4]: | |||
| score1 = 0.10 | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| else: | |||
| score1 = float(abs(rect[2] - list2[j][4]) / 90) | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| if min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1]) > traffic_dict['roundness']: | |||
| score2 = (min(rect[1][0], rect[1][1]) - max(rect[1][0], rect[1][1]) * | |||
| traffic_dict['roundness']) / ( | |||
| max(rect[1][0], rect[1][1]) * ( | |||
| 1 - traffic_dict['roundness'])) | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| elif abs(rect[2] - list2[j][4]) > traffic_dict['speedRoadVehicleAngleMax'] and abs( | |||
| rect[2] - list2[j][4]) <= 90: | |||
| if rect[1][0] >= rect[1][1] or ( | |||
| min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1])) >= \ | |||
| traffic_dict['roundness']: | |||
| score1 = float(abs(rect[2] - list2[i][4]) / 90) | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| if min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1]) > traffic_dict[ | |||
| 'roundness']: | |||
| score2 = (min(rect[1][0], rect[1][1]) - max(rect[1][0], rect[1][1]) * | |||
| traffic_dict['roundness']) / ( | |||
| max(rect[1][0], rect[1][1]) * ( | |||
| 1 - traffic_dict['roundness'])) | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| break | |||
| else: | |||
| j += 1 | |||
| flag = True | |||
| if flag == True: | |||
| break | |||
| i += 1 | |||
| list22 = [] | |||
| a = 0 | |||
| for a in range(len(list4)): | |||
| tmp = list4[a] | |||
| list4[a] = list4[0] | |||
| list4[0] = tmp | |||
| for j in range(1, len(list4)): | |||
| for k in range(4): | |||
| point1 = [list4[0][k][0], list4[0][k][1]] | |||
| for w in range(3): | |||
| line1 = [list4[j][w][0], list4[j][w][1], list4[j][w + 1][0], list4[j][w + 1][1]] | |||
| list22.append(point_to_line_distance(point1, line1)) | |||
| w += 1 | |||
| line1 = [list4[j][3][0], list4[j][3][1], list4[j][0][0], list4[j][0][1]] | |||
| list22.append(point_to_line_distance(point1, line1)) | |||
| k += 1 | |||
| j += 1 | |||
| traffic_dict['det'][a][8] = min(list22) | |||
| list22 = [] | |||
| if traffic_dict['det'][a][8] <= list3[a] * traffic_dict['vehicleFactor']: | |||
| score1 = 1 - traffic_dict['det'][a][8] / (list3[a] * traffic_dict['vehicleFactor']) | |||
| traffic_dict['det'][a][9][2] = score1 | |||
| if max(traffic_dict['det'][a][9]) == traffic_dict['det'][a][9][0] and traffic_dict['det'][a][9][0] != -1: | |||
| traffic_dict['det'][a][10] = 0 | |||
| elif max(traffic_dict['det'][a][9]) == traffic_dict['det'][a][9][1] and traffic_dict['det'][a][9][1] != -1: | |||
| traffic_dict['det'][a][10] = 1 | |||
| elif max(traffic_dict['det'][a][9]) == traffic_dict['det'][a][9][2] and traffic_dict['det'][a][9][2] != -1: | |||
| traffic_dict['det'][a][10] = 2 | |||
| else: | |||
| traffic_dict['det'][a][10] = 3 | |||
| a += 1 | |||
| # print(traffic_dict['det']) # [[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, [角度得分, 长宽比得分, 最小距离得分], 类别], ...] | |||
| list23 = traffic_dict['det'] | |||
| traffic_dict['det'] = [] | |||
| for i in range(len(list23)): | |||
| list23[i][9] = max(list23[i][9]) | |||
| #print("line393", list23) # 目标对象, [[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, max([角度得分, 长宽比得分, 最小距离得分]), 类别], ...] | |||
| t6 = time.time() | |||
| time_infos = 'postTime:%.2f (分割时间：%.2f, findContours:%.2f ruleJudge:%.2f)' % ( | |||
| get_ms(t6, t3), get_ms(t4, t3), get_ms(t5, t4), get_ms(t6, t5)) | |||
| return list23, image, time_infos | |||
| def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| """ | |||
| 对于字典traffic_dict中的各个键，说明如下： | |||
| speedRoadArea：speedRoad的最小外接矩形的面积 | |||
| speedRoadVehicleAngleMin：判定发生交通事故的speedRoad与vehicle间的最小夹角 | |||
| vehicleCoordinate：是一个列表，用于存储被检测出的vehicle的坐标（vehicle检测模型） | |||
| roundness：圆度 ,vehicle的长与宽的比率，设置为0.7，若宽与长的比值大于0.7，则判定该vehicle发生交通事故 | |||
| ZoomFactor：存储的是图像在H和W方向上的缩放因子，其值小于1 | |||
| 'cls':类别号 | |||
| 'vehicleFactor':两辆车之间的安全距离被定义为：min(车辆1的宽，车辆2的宽) * vehicleFactor | |||
| 未发生交通事故时，得分为-1，”事故类型“为3 | |||
| 最终输出格式：[[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, max([角度得分, 长宽比得分, 最小距离得分]), 交通事故类别], ...] | |||
| 交通事故类别：0表示角度，1表示长宽比，2表示最短距离，3表示未发生交通事故 | |||
| """ | |||
| det_cors = [] | |||
| for bb in traffic_dict['det']: | |||
| det_cors.append((int(bb[1]), int(bb[2]))) | |||
| @@ -674,18 +913,24 @@ def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| traffic_dict['vehicleCoordinate'] = det_cors | |||
| t3 = time.time() | |||
| list17 = [] | |||
| list21 = [] # 存储一副图像中vehicles的contours | |||
| list8 = [] | |||
| list10 = [] # 存储一副图像中vehicles的contours | |||
| image_speedRoad = traffic_dict['mask'].copy() | |||
| image_vehicle = traffic_dict['mask'].copy() | |||
| image_speedRoad[image_speedRoad == 2] = 0 # 将vehicle过滤掉，只包含背景和speedRoad | |||
| image_vehicle[image_vehicle == 1] = 0 # 将speedRoad过滤掉，只包含背景和vehicle | |||
| image_speedRoad = cv2.cvtColor(np.uint8(image_speedRoad), cv2.COLOR_RGB2BGR) # 道路 | |||
| image_vehicle = cv2.cvtColor(np.uint8(image_vehicle), cv2.COLOR_RGB2BGR) # 车辆 | |||
| # image_vehicle = cv2.resize(image_vehicle, (3840, 2160)) | |||
| # cv2.imwrite('./demo/vehicle.png', image_vehicle) | |||
| # image_speedRoad = cv2.resize(image_speedRoad, (1920, 1080)) | |||
| # cv2.imwrite('./demo/speedRoad.png', image_speedRoad) | |||
| if debug == False: | |||
| for key in traffic_dict['label_info']: | |||
| list17.append(traffic_dict['label_info'][key]) | |||
| colour_codes = np.array(list17) # [[0 0 0],[128 0 0],[0 128 0]] ndarray类型 | |||
| list8.append(traffic_dict['label_info'][key]) | |||
| colour_codes = np.array(list8) # [[0 0 0],[128 0 0],[0 128 0]] ndarray类型 | |||
| preds_squeeze_predict = colour_codes[traffic_dict['mask']] | |||
| image = cv2.cvtColor(np.uint8(preds_squeeze_predict), cv2.COLOR_RGB2BGR) | |||
| else: | |||
| @@ -693,13 +938,14 @@ def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| t4 = time.time() | |||
| img1 = cv2.cvtColor(image_speedRoad, cv2.COLOR_BGR2GRAY) | |||
| contours, hierarchy = cv2.findContours(img1, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) | |||
| # contours, hierarchy = cv2.findContours(img1, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) | |||
| contours, hierarchy = cv2.findContours(img1, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) | |||
| t5 = time.time() | |||
| list1 = [] # 过渡使用 | |||
| list2 = [] # 存放道路坐标(Xmin,Xmax,Ymin,Ymax) | |||
| list3 = [] # # 存储vehicle最小外接矩形的最短边 | |||
| list4 = [] # list4存储车辆的box参数 | |||
| list7 = [] # 存储vehicle的宽高 | |||
| list5 = [] # 存储vehicle的宽高 | |||
| for cnt in contours: # 道路 | |||
| rect = cv2.minAreaRect(cnt) | |||
| if rect[1][0] * rect[1][1] > traffic_dict['speedRoadArea']: # 过滤掉面积小于阈值的speedRoad | |||
| @@ -712,11 +958,21 @@ def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| list1.append(Xmax) | |||
| list1.append(Ymin) | |||
| list1.append(Ymax) | |||
| list1.append(rect[2]) # 将道路矩形框与水平方向的夹角存储在list1中 | |||
| if rect[1][0] <= rect[1][1]: | |||
| if rect[2] >= 0 and rect[2] < 90: | |||
| speedRoadAngle = rect[2] + 90 | |||
| elif rect[2] == 90: | |||
| speedRoadAngle = 0 | |||
| elif rect[2] >= 0 and rect[2] <= 90: | |||
| speedRoadAngle = rect[2] | |||
| list1.append(speedRoadAngle) # 将道路矩形框与水平方向的夹角存储在list1中 | |||
| list1.append(rect[1]) # 将道路的宽高存储在list1中 | |||
| list2.append(list1) | |||
| list1 = [] | |||
| # print("line203", list2) | |||
| # print("line213", list2) | |||
| list14 = [] | |||
| list15 = [] | |||
| count = 0 | |||
| for i in range(0, len(traffic_dict['vehicleCoordinate']), 2): | |||
| mask = np.zeros(image_vehicle.shape[:2], dtype="uint8") | |||
| @@ -727,100 +983,100 @@ def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| cv2.rectangle(mask, (x0, y0), (x1, y1), 255, -1, lineType=cv2.LINE_AA) | |||
| image_vehicle_masked = cv2.bitwise_and(image_vehicle, image_vehicle, mask=mask) | |||
| # cv2.imwrite('./demo/vehicle_masked.png', image_vehicle_masked) | |||
| # cv2.imwrite('./demo/vehicle_masked_' + str(i) + '.png', cv2.resize(image_vehicle_masked, (3840, 2160))) | |||
| # cv2.imwrite('./demo/vehicle_masked_' + str(i) + '.png', cv2.resize(image_vehicle_masked, (1920, 1080))) | |||
| img2 = cv2.cvtColor(image_vehicle_masked, cv2.COLOR_BGR2GRAY) | |||
| contours2, hierarchy2 = cv2.findContours(img2, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) | |||
| if len(contours2) != 0: | |||
| list21.append(contours2) | |||
| # contours2, hierarchy2 = cv2.findContours(img2, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) | |||
| contours2, hierarchy2 = cv2.findContours(img2, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) | |||
| list24 = [] | |||
| if len(contours2) != 0: | |||
| list10.append(contours2) | |||
| list14.append(traffic_dict['det'][count]) | |||
| else: | |||
| traffic_dict['det'][i].append(5) | |||
| traffic_dict['det'][i].append(0.5) | |||
| traffic_dict['det'][i].append(15) | |||
| traffic_dict['det'][i].append(0.01) | |||
| traffic_dict['det'][i].append(3) | |||
| list15.append(traffic_dict['det'][i]) | |||
| count += 1 | |||
| traffic_dict['det'] = list14 | |||
| # list12 = [] | |||
| # print("line222", len(list21)) | |||
| if len(list21) != 0: | |||
| for i in range(len(list21)): # 选取落在检测框范围内的分割区域，存在一个问题，就是其他vehicle的一小部分分割区域可能也落在了检测框中，这时会产生多个contours， | |||
| if len(list21[i]) > 1: # 这里我通过比较同一检测框内各个contours对应的最小外接矩形的面积，来剔除那些存在干扰的contours，最终只保留一个contours | |||
| for j in range(len(list21[i])): | |||
| contours_j = list21[i][j] | |||
| if len(list10) != 0: | |||
| for i in range(len(list10)): # 选取落在检测框范围内的分割区域，存在一个问题，就是其他vehicle的一小部分分割区域可能也落在了检测框中，这时会产生多个contours， | |||
| list12 = [] | |||
| if len(list10[i]) > 1: # 这里我通过比较同一检测框内各个contours对应的最小外接矩形的面积，来剔除那些存在干扰的contours，最终只保留一个contours | |||
| for j in range(len(list10[i])): | |||
| contours_j = list10[i][j] | |||
| rect = cv2.minAreaRect(contours_j) | |||
| list24.append(rect) | |||
| list25 = [list24[g][1][0] * list24[g][1][1] for g in range(len(list24))] | |||
| maxAreaIndex = list25.index(max(list25)) | |||
| rect = list24[maxAreaIndex] | |||
| list12.append(rect) | |||
| # print("line257", rect) | |||
| # print("line258", len(list10[i]), len(list12)) | |||
| list13 = [list12[g][1][0] * list12[g][1][1] for g in range(len(list12))] | |||
| maxAreaIndex = list13.index(max(list13)) | |||
| rect = list12[maxAreaIndex] | |||
| box = cv2.boxPoints(rect).astype(np.int32) | |||
| elif len(list21[i]) == 1: | |||
| rect = cv2.minAreaRect(np.array(list21[i][0])) | |||
| # print("line261", list10[i]) | |||
| # print("line262", list13) | |||
| # print("line263", maxAreaIndex) | |||
| maxAreacontours = list10[i][maxAreaIndex] | |||
| ellipse = cv2.fitEllipse(maxAreacontours) # （中心点坐标(x，y)，(短轴长度，长轴长度)，短轴与水平线的夹角） | |||
| if ellipse[2] >= 0 and ellipse[2] < 90: | |||
| vehicleAngle = 90 + ellipse[2] | |||
| elif ellipse[2] >= 90 and ellipse[2] < 180: | |||
| vehicleAngle = ellipse[2] - 90 | |||
| elif ellipse[2] == 180: | |||
| vehicleAngle = 90 | |||
| elif len(list10[i]) == 1: | |||
| rect = cv2.minAreaRect(np.array(list10[i][0])) | |||
| box = cv2.boxPoints(rect).astype(np.int32) | |||
| maxAreacontours = list10[i][0] | |||
| ellipse = cv2.fitEllipse(maxAreacontours) | |||
| if ellipse[2] >= 0 and ellipse[2] < 90: | |||
| vehicleAngle = 90 + ellipse[2] | |||
| elif ellipse[2] >= 90 and ellipse[2] < 180: | |||
| vehicleAngle = ellipse[2] - 90 | |||
| elif ellipse[2] == 180: | |||
| vehicleAngle = 90 | |||
| list3.append(min(rect[1])) | |||
| list4.append(box) | |||
| list7.append(rect[1]) | |||
| list5.append(rect[1]) | |||
| for j in range(len(list2)): | |||
| if (rect[0][0] > list2[j][0] and rect[0][0] < list2[j][1]) and (rect[0][1] > list2[j][2] and rect[0][1] < list2[j][3]): # 判断车辆矩形框的中心点坐标是否在道路矩形框Xmin,Xmax,Ymin和Ymax的范围内； | |||
| box = cv2.boxPoints(rect).astype(np.int32) # 将Box2D结构作为输入并返回4个角点。 | |||
| if (box[0][0] >= list2[j][0] and box[0][0] <= list2[j][1] and box[0][1] >= list2[j][2] and box[0][1] <= list2[j][3]) and (box[1][0] >= list2[j][0] and box[1][0] <= list2[j][1] and box[1][1] >= list2[j][2] and box[1][1] <= list2[j][3]) and (box[2][0] >= list2[j][0] and box[2][0] <= list2[j][1] and box[2][1] >= list2[j][2] and box[2][1] <= list2[j][3]) and (box[3][0] >= list2[j][0] and box[3][0] <= list2[j][1] and box[3][1] >= list2[j][2] and box[3][1] <= list2[j][3]): # 比较车辆矩形框四个顶点的坐标是否都在道路矩形框Xmin,Xmax,Ymin和Ymax的范围内，若都在，则说明该车辆在这条道路上。 | |||
| y_min = min(box[0][1], box[1][1], box[2][1], box[3][1]) | |||
| y_max = max(box[0][1], box[1][1], box[2][1], box[3][1]) | |||
| angle = abs(rect[2] - list2[j][4]) | |||
| if list2[j][4] == 0: | |||
| list2[j][4] = 90 | |||
| speedRoadVehicleAngle = abs(vehicleAngle - list2[j][4]) | |||
| # print("line288", vehicleAngle, list2[j][4], speedRoadVehicleAngle) | |||
| roundness = min(rect[1][0], rect[1][1]) / max(rect[1][0], rect[1][1]) | |||
| traffic_dict['det'][i].append(angle) | |||
| traffic_dict['det'][i].append(speedRoadVehicleAngle) | |||
| traffic_dict['det'][i].append(roundness) | |||
| traffic_dict['det'][i].append(999) # 给最小距离占位 | |||
| traffic_dict['det'][i].append([-1, -1, -1]) | |||
| traffic_dict['det'][i].append(666) # 给事故类别占位 | |||
| score1 = float(abs(rect[2] - list2[j][4]) / 90) | |||
| score2 = (min(rect[1][0], rect[1][1]) - max(rect[1][0], rect[1][1]) * traffic_dict[ | |||
| 'roundness']) / (max(rect[1][0], rect[1][1]) * ( | |||
| 1 - traffic_dict['roundness'])) | |||
| if abs(rect[2] - list2[j][4]) >= traffic_dict['speedRoadVehicleAngleMin'] and abs( | |||
| rect[2] - list2[j][4]) <= traffic_dict[ | |||
| 'speedRoadVehicleAngleMax']: # 当道路同水平方向的夹角与车辆同水平方向的夹角的差值在15°和75°之间时，需要将车辆框出来 | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| if y_min > 0 and y_max < image_vehicle.shape[0]: # 过滤掉上下方被speedRoad的边界截断的vehicle | |||
| if speedRoadVehicleAngle >= traffic_dict['speedRoadVehicleAngleMin']: # 当道路同水平方向的夹角与车辆同水平方向的夹角的差值在15°和75°之间时，需要将车辆框出来 | |||
| if speedRoadVehicleAngle > 90: | |||
| score1 = float((180 - speedRoadVehicleAngle) / 90) | |||
| else: | |||
| score1 = float(speedRoadVehicleAngle / 90) | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| if roundness > traffic_dict['roundness']: | |||
| score2 = (min(rect[1][0], rect[1][1]) - max(rect[1][0], rect[1][1]) * traffic_dict[ | |||
| 'roundness']) / (max(rect[1][0], rect[1][1]) * (1 - traffic_dict['roundness'])) | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| elif list2[j][5][0] < list2[j][5][1]: # speedRoad的最小外接矩形的w<h，说明该speedRoad是左侧道路 | |||
| if y_min > 0 and y_max < image_vehicle.shape[0]: # 过滤掉上下方被speedRoad的边界截断的vehicle | |||
| if abs(rect[2] - list2[j][4]) >= 0 and abs(rect[2] - list2[j][4]) < traffic_dict['speedRoadVehicleAngleMin']: | |||
| if rect[1][0] >= rect[1][1]: | |||
| if score1 != 0: | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| else: | |||
| traffic_dict['det'][i][9][0] = 1 | |||
| if roundness > traffic_dict['roundness']: | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| elif abs(rect[2] - list2[j][4]) > traffic_dict['speedRoadVehicleAngleMax'] and abs( | |||
| rect[2] - list2[j][4]) <= 90: | |||
| if rect[1][0] <= rect[1][1]: | |||
| if score1 != 0: | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| else: | |||
| traffic_dict['det'][i][9][0] = 1 | |||
| if roundness > traffic_dict['roundness']: | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| elif list2[j][5][0] > list2[j][5][1]: # speedRoad的最小外接矩形的w>h，说明该speedRoad是右侧道路 | |||
| if y_min > 0 and y_max < image_vehicle.shape[0]: | |||
| if abs(rect[2] - list2[j][4]) >= 0 and abs(rect[2] - list2[j][4]) < traffic_dict['speedRoadVehicleAngleMin']: | |||
| if rect[1][0] <= rect[1][1]: | |||
| if score1 != 0: | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| else: | |||
| traffic_dict['det'][i][9][0] = 1 | |||
| if roundness > traffic_dict['roundness']: | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| elif abs(rect[2] - list2[j][4]) > traffic_dict['speedRoadVehicleAngleMax'] and abs( | |||
| rect[2] - list2[j][4]) <= 90: | |||
| if rect[1][0] >= rect[1][1]: | |||
| if score1 != 0: | |||
| traffic_dict['det'][i][9][0] = score1 | |||
| else: | |||
| traffic_dict['det'][i][9][0] = 1 | |||
| if roundness > traffic_dict['roundness']: | |||
| traffic_dict['det'][i][9][1] = score2 | |||
| break | |||
| else: | |||
| j += 1 | |||
| i += 1 | |||
| list22 = [] | |||
| list11 = [] | |||
| if len(list4) > 1: | |||
| for a in range(len(list4)): | |||
| tmp = list4[a] | |||
| @@ -831,19 +1087,19 @@ def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| point1 = [list4[0][k][0], list4[0][k][1]] | |||
| for w in range(3): | |||
| line1 = [list4[j][w][0], list4[j][w][1], list4[j][w + 1][0], list4[j][w + 1][1]] | |||
| list22.append(point_to_line_distance(point1, line1)) | |||
| list11.append(point_to_line_distance(point1, line1)) | |||
| w += 1 | |||
| line1 = [list4[j][3][0], list4[j][3][1], list4[j][0][0], list4[j][0][1]] | |||
| list22.append(point_to_line_distance(point1, line1)) | |||
| list11.append(point_to_line_distance(point1, line1)) | |||
| k += 1 | |||
| j += 1 | |||
| traffic_dict['det'][a][8] = min(list22) | |||
| list22 = [] | |||
| traffic_dict['det'][a][8] = min(list11) | |||
| list11 = [] | |||
| if traffic_dict['det'][a][8] < list3[a] * traffic_dict['vehicleFactor']: | |||
| score1 = 1 - traffic_dict['det'][a][8] / (list3[a] * traffic_dict['vehicleFactor']) | |||
| traffic_dict['det'][a][9][2] = score1 | |||
| # print("line334", traffic_dict['det']) # [[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, [角度得分, 长宽比得分, 最小距离得分], 类别], ...] | |||
| if max(traffic_dict['det'][a][9]) == traffic_dict['det'][a][9][0] and traffic_dict['det'][a][9][0] != -1: | |||
| traffic_dict['det'][a][10] = 0 | |||
| elif max(traffic_dict['det'][a][9]) == traffic_dict['det'][a][9][1] and traffic_dict['det'][a][9][1] != -1: | |||
| @@ -860,18 +1116,13 @@ def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| traffic_dict['det'][0][10] = 1 | |||
| else: | |||
| traffic_dict['det'][0][10] = 3 | |||
| # print("line347", traffic_dict['det']) # [[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, [角度得分, 长宽比得分, 最小距离得分], 类别], ...] | |||
| # print("line351", traffic_dict['det']) # [[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, [角度得分, 长宽比得分, 最小距离得分], 类别], ...] | |||
| list23 = traffic_dict['det'] | |||
| for i in range(len(list23)): | |||
| if len(list23[i]) != 11: | |||
| list23[i].append(90) | |||
| list23[i].append(0.7) | |||
| list23[i].append(10) | |||
| list23[i].append(1) | |||
| list23[i].append(2) | |||
| else: | |||
| list23[i][9] = max(list23[i][9]) | |||
| # print("line351", list23) # 目标对象, [[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, max([角度得分, 长宽比得分, 最小距离得分]), 类别], ...] | |||
| list23[i][9] = max(list23[i][9]) | |||
| for i in range(len(list15)): | |||
| list23.append(list15[i]) | |||
| # print("line357", list23) # 目标对象, [[cls, x0, y0, x1, y1, score, 角度, 长宽比, 最小距离, max([角度得分, 长宽比得分, 最小距离得分]), 类别], ...] | |||
| else: | |||
| print("分割模型未检测到vehicle！") | |||
| list23 = traffic_dict['det'] | |||
| @@ -879,4 +1130,5 @@ def trafficPostProcessingV2(traffic_dict, debug=False): | |||
| t6 = time.time() | |||
| time_infos = 'postTime:%.2f (分割时间：%.2f, findContours:%.2f ruleJudge:%.2f)' % ( | |||
| get_ms(t6, t3), get_ms(t4, t3), get_ms(t5, t4), get_ms(t6, t5)) | |||
| return list23, image, time_infos | |||
| return list23, image, time_infos | |||
Notiek ielāde…