traffic_counter.zip_opencv_视频车辆计数_计数_车辆检测_车辆计数

import numpy as np import cv2 import pandas as pd cap = cv2.VideoCapture('traffic.mp4') frames_count, fps, width, height = cap.get(cv2.CAP_PROP_FRAME_COUNT), cap.get(cv2.CAP_PROP_FPS), cap.get( cv2.CAP_PROP_FRAME_WIDTH), cap.get(cv2.CAP_PROP_FRAME_HEIGHT) width = int(width) height = int(height) print(frames_count, fps, width, height) # creates a pandas data frame with the number of rows the same length as frame count df = pd.DataFrame(index=range(int(frames_count))) df.index.name = "Frames" framenumber = 0 # keeps track of current frame carscrossedup = 0 # keeps track of cars that crossed up carscrosseddown = 0 # keeps track of cars that crossed down carids = [] # blank list to add car ids caridscrossed = [] # blank list to add car ids that have crossed totalcars = 0 # keeps track of total cars fgbg = cv2.createBackgroundSubtractorMOG2() # create background subtractor # information to start saving a video file ret, frame = cap.read() # import image ratio = .5 # resize ratio image = cv2.resize(frame, (0, 0), None, ratio, ratio) # resize image width2, height2, channels = image.shape video = cv2.VideoWriter('traffic_counter.avi', cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), fps, (height2, width2), 1) while True: ret, frame = cap.read() # import image if ret: # if there is a frame continue with code image = cv2.resize(frame, (0, 0), None, ratio, ratio) # resize image gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) # converts image to gray fgmask = fgbg.apply(gray) # uses the background subtraction # applies different thresholds to fgmask to try and isolate cars # just have to keep playing around with settings until cars are easily identifiable kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5)) # kernel to apply to the morphology closing = cv2.morphologyEx(fgmask, cv2.MORPH_CLOSE, kernel) opening = cv2.morphologyEx(closing, cv2.MORPH_OPEN, kernel) dilation = cv2.dilate(opening, kernel) retvalbin, bins = cv2.threshold(dilation, 220, 255, cv2.THRESH_BINARY) # removes the shadows # creates contours im2, contours, hierarchy = cv2.findContours(bins, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) # use convex hull to create polygon around contours hull = [cv2.convexHull(c) for c in contours] # draw contours cv2.drawContours(image, hull, -1, (0, 255, 0), 3) # line created to stop counting contours, needed as cars in distance become one big contour lineypos = 225 cv2.line(image, (0, lineypos), (width, lineypos), (255, 0, 0), 5) # line y position created to count contours lineypos2 = 250 cv2.line(image, (0, lineypos2), (width, lineypos2), (0, 255, 0), 5) # min area for contours in case a bunch of small noise contours are created minarea = 300 # max area for contours, can be quite large for buses maxarea = 50000 # vectors for the x and y locations of contour centroids in current frame cxx = np.zeros(len(contours)) cyy = np.zeros(len(contours)) for i in range(len(contours)): # cycles through all contours in current frame if hierarchy[0, i, 3] == -1: # using hierarchy to only count parent contours (contours not within others) area = cv2.contourArea(contours[i]) # area of contour if minarea < area < maxarea: # area threshold for contour # calculating centroids of contours cnt = contours[i] M = cv2.moments(cnt) cx = int(M['m10'] / M['m00']) cy = int(M['m01'] / M['m00']) if cy > lineypos: # filters out contours that are above line (y starts at top) # gets bounding points of contour to create rectangle # x,y is top left corner and w,h is width and height x, y, w, h = cv2.boundingRect(cnt) # creates a rectangle around contour cv2.rectangle(image, (x, y), (x + w, y + h), (255, 0, 0), 2) # Prints centroid text in order to double check later on cv2.putText(image, str(cx) + "," + str(cy), (cx + 10, cy + 10), cv2.FONT_HERSHEY_SIMPLEX, .3, (0, 0, 255), 1) cv2.drawMarker(image, (cx, cy), (0, 0, 255), cv2.MARKER_STAR, markerSize=5, thickness=1, line_type=cv2.LINE_AA) # adds centroids that passed previous criteria to centroid list cxx[i] = cx cyy[i] = cy # eliminates zero entries (centroids that were not added) cxx = cxx[cxx != 0] cyy = cyy[cyy != 0] # empty list to later check which centroid indices were added to dataframe minx_index2 = [] miny_index2 = [] # maximum allowable radius for current frame centroid to be considered the same centroid from previous frame maxrad = 25 # The section below keeps track of the centroids and assigns them to old carids or new carids if len(cxx): # if there are centroids in the specified area if not carids: # if carids is empty for i in range(len(cxx)): # loops through all centroids carids.append(i) # adds a car id to the empty list carids df[str(carids[i])] = "" # adds a column to the dataframe corresponding to a carid # assigns the centroid values to the current frame (row) and carid (column) df.at[int(framenumber), str(carids[i])] = [cxx[i], cyy[i]] totalcars = carids[i] + 1 # adds one count to total cars else: # if there are already car ids dx = np.zeros((len(cxx), len(carids))) # new arrays to calculate deltas dy = np.zeros((len(cyy), len(carids))) # new arrays to calculate deltas for i in range(len(cxx)): # loops through all centroids for j in range(len(carids)): # loops through all recorded car ids # acquires centroid from previous frame for specific carid oldcxcy = df.iloc[int(framenumber - 1)][str(carids[j])] # acquires current frame centroid that doesn't necessarily line up with previous frame centroid curcxcy = np.array([cxx[i], cyy[i]]) if not oldcxcy: # checks if old centroid is empty in case car leaves screen and new car shows continue # continue to next carid else: # calculate centroid deltas to compare to current frame position later dx[i, j] = oldcxcy[0] - curcxcy[0] dy[i, j] = oldcxcy[1] - curcxcy[1] for j in range(len(carids)): # loops through all current car ids sumsum = np.abs(dx[:, j]) + np.abs(dy[:, j]) # sums the deltas wrt to car ids # finds which index carid had the min difference and this is true index correctindextrue = np.argmin(np.abs(sumsum)) minx_index = correctindextrue miny_index = correctindextrue # acquires delta values of the minimum deltas in order to check if it is within radius later on mindx = dx[minx_index, j] mindy = dy[miny_index, j] if mindx == 0 and mindy == 0 and np.all(dx[:, j] == 0) and np.all(dy[:, j] == 0): # checks if minimum value is 0 and checks if all deltas are zero since this is empty set # delta could be zero