python+opencv車道線檢測(cè)（簡(jiǎn)易實(shí)現(xiàn)），供大家參考，具體內(nèi)容如下

技術(shù)棧：python+opencv

1、canny邊緣檢測(cè)獲取圖中的邊緣信息；2、霍夫變換尋找圖中直線；3、繪制梯形感興趣區(qū)域獲得車前范圍；4、得到并繪制車道線；

import cv2import numpy as npdef canny(): gray = cv2.cvtColor(lane_image, cv2.COLOR_RGB2GRAY) #高斯濾波 blur = cv2.GaussianBlur(gray, (5, 5), 0) #邊緣檢測(cè) canny_img = cv2.Canny(blur, 50, 150) return canny_imgdef region_of_interest(r_image): h = r_image.shape[0] w = r_image.shape[1] # 這個(gè)區(qū)域不穩(wěn)定，需要根據(jù)圖片更換 poly = np.array([ [(100, h), (500, h), (290, 180), (250, 180)] ]) mask = np.zeros_like(r_image) # 繪制掩膜圖像 cv2.fillPoly(mask, poly, 255) # 獲得ROI區(qū)域 masked_image = cv2.bitwise_and(r_image, mask) return masked_imageif __name__ == ’__main__’: image = cv2.imread(’test.jpg’) lane_image = np.copy(image) canny = canny() cropped_image = region_of_interest(canny) cv2.imshow('result', cropped_image) cv2.waitKey(0)霍夫變換加線性擬合改良：

效果圖：

代碼實(shí)現(xiàn)：

主要增加了根據(jù)斜率作線性擬合過濾無用點(diǎn)后連線的操作；

import cv2import numpy as npdef canny(): gray = cv2.cvtColor(lane_image, cv2.COLOR_RGB2GRAY) blur = cv2.GaussianBlur(gray, (5, 5), 0) canny_img = cv2.Canny(blur, 50, 150) return canny_imgdef region_of_interest(r_image): h = r_image.shape[0] w = r_image.shape[1] poly = np.array([ [(100, h), (500, h), (280, 180), (250, 180)] ]) mask = np.zeros_like(r_image) cv2.fillPoly(mask, poly, 255) masked_image = cv2.bitwise_and(r_image, mask) return masked_imagedef get_lines(img_lines): if img_lines is not None: for line in lines: for x1, y1, x2, y2 in line: # 分左右車道 k = (y2 - y1) / (x2 - x1) if k < 0: lefts.append(line) else: rights.append(line)def choose_lines(after_lines, slo_th): # 過濾斜率差別較大的點(diǎn) slope = [(y2 - y1) / (x2 - x1) for line in after_lines for x1, x2, y1, y2 in line] # 獲得斜率數(shù)組 while len(after_lines) > 0: mean = np.mean(slope) # 計(jì)算平均斜率 diff = [abs(s - mean) for s in slope] # 每條線斜率與平均斜率的差距 idx = np.argmax(diff) # 找到最大斜率的索引 if diff[idx] > slo_th: # 大于預(yù)設(shè)的閾值選取 slope.pop(idx) after_lines.pop(idx) else: break return after_linesdef clac_edgepoints(points, y_min, y_max): x = [p[0] for p in points] y = [p[1] for p in points] k = np.polyfit(y, x, 1) # 曲線擬合的函數(shù)，找到xy的擬合關(guān)系斜率 func = np.poly1d(k) # 斜率代入可以得到一個(gè)y=kx的函數(shù) x_min = int(func(y_min)) # y_min = 325其實(shí)是近似找了一個(gè) x_max = int(func(y_max)) return [(x_min, y_min), (x_max, y_max)]if __name__ == ’__main__’: image = cv2.imread(’F:A_javaProtest.jpg’) lane_image = np.copy(image) canny_img = canny() cropped_image = region_of_interest(canny_img) lefts = [] rights = [] lines = cv2.HoughLinesP(cropped_image, 1, np.pi / 180, 15, np.array([]), minLineLength=40, maxLineGap=20) get_lines(lines) # 分別得到左右車道線的圖片 good_leftlines = choose_lines(lefts, 0.1) # 處理后的點(diǎn) good_rightlines = choose_lines(rights, 0.1) leftpoints = [(x1, y1) for left in good_leftlines for x1, y1, x2, y2 in left] leftpoints = leftpoints + [(x2, y2) for left in good_leftlines for x1, y1, x2, y2 in left] rightpoints = [(x1, y1) for right in good_rightlines for x1, y1, x2, y2 in right] rightpoints = rightpoints + [(x2, y2) for right in good_rightlines for x1, y1, x2, y2 in right] lefttop = clac_edgepoints(leftpoints, 180, image.shape[0]) # 要畫左右車道線的端點(diǎn) righttop = clac_edgepoints(rightpoints, 180, image.shape[0]) src = np.zeros_like(image) cv2.line(src, lefttop[0], lefttop[1], (255, 255, 0), 7) cv2.line(src, righttop[0], righttop[1], (255, 255, 0), 7) cv2.imshow(’line Image’, src) src_2 = cv2.addWeighted(image, 0.8, src, 1, 0) cv2.imshow(’Finally Image’, src_2) cv2.waitKey(0)

待改進(jìn)：

代碼實(shí)用性差，幾乎不能用于實(shí)際，但是可以作為初學(xué)者的練手項(xiàng)目；斑馬線檢測(cè)思路：獲取車前感興趣區(qū)域，判斷白色像素點(diǎn)比例即可實(shí)現(xiàn)；行人檢測(cè)思路：opencv有內(nèi)置行人檢測(cè)函數(shù)，基于內(nèi)置的訓(xùn)練好的數(shù)據(jù)集；

以上就是本文的全部?jī)?nèi)容，希望對(duì)大家的學(xué)習(xí)有所幫助，也希望大家多多支持好吧啦網(wǎng)。