opencv小项目——车道检测

对于视频，只要处理好每一帧的车道检测，整个视频的车道检测也相应处理成功，因此，先处理一帧的图像

步骤1.灰度化、滤波

gray = cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY)
# 采用高斯滤波
blur = cv2.GaussianBlur(gray,(5,5),0)

步骤2.边缘检测

canny = cv2.Canny(blur,50,150)

步骤3.不规则ROI截取

height,width = canny.shape[0]
# 端点根据实际测量实验得出
points = np.array([[(200, height),(800, 300),(1200, height),]],np.int32)
mask = np.zeros_like(canny)
# 产生的多边形应包含全部所需车道，但也应尽可能小，以减少噪声
cv2.fillPoly(mask,points,(255,255,255))
img = cv2.bitwise_and(canny,mask)

步骤4.霍夫直线检测

# lines包含检测到的直接的端点的坐标，shape=(n,1,4)
lines = cv2.HoughLinesP(img,1,np.pi/180,50,minLineLength=30,maxLineGap=20)

步骤5.分离左右车道

# 根据斜率的正负来分离左右车道
left_lines = []
right_lines = []
for line in lines:
  for x1,y1,x2,y2 in line:
    k = (y2-y1)/(x2-x1)
    if k>0:
      right_lines.append(line)
    else:
      left_lines.append(line)

步骤6.排除异常数据

# 利用各直线斜率和斜率均值的差，排除超过阈值的异常直线
def clean_lines(lines,threshold):
    ks = [(y2-y1)/(x2-x1) for line in lines for x1,y1,x2,y2 in line]
    mean = np.mean(ks)
    while len(lines) > 0:
        diff = [abs(k-mean) for k in ks]
        idx = np.argmax(diff)
        if diff[idx] > threshold:
            lines.pop(idx)
            ks.pop(idx)
        else:
            break
            
clean_lines(left_lines,0.1)
clean_lines(right_lines,0.1)

步骤7.最小二乘法拟合左右车道线

def least_squares_fit(lines,ymin,ymax):
    x = [x1 for line in lines for x1,y1,x2,y2 in line]
    y = [y1 for line in lines for x1,y1,x2,y2 in line]
    x += [x2 for line in lines for x1,y1,x2,y2 in line]
    y += [y2 for line in lines for x1,y1,x2,y2 in line]
    
    # 拟合函数
    fit_fn = np.poly1d(np.polyfit(y,x,1))
    return [(int(fit_fn(ymin)),ymin),(int(fit_fn(ymax)),ymax)]
  
left = least_squares_fit(left_lines,325,height)
right = least_squares_fit(right_lines,325,height)

步骤8.画左右车道线

cv2.line(frame,left[0],left[1],(0,255,0),3)
cv2.line(frame,right[0],right[1],(0,255,0),3)

步骤9.对视频的每一帧应用上述步骤

qwq,这个自食其力吧

项目的缺陷

1. 只能对直线车道有很好的识别率，在弯道时正确检测率很低

2. 需要人工测量修改ROI，若ROI中有多条车道，则结果偏离预期

3. 车道线缺失大部分时，检测不出该边车道

   

4. 项目鲁棒性不够，当画面中没有车道时会中断退出

5. 车道只有黄色和白色，因此可以分离出黄色和白色，从而可以减少误差