图形识别 智能驾驶

车辆工程 智能驾驶

3 人赞同了该文章

图形识别也可以理解为轮廓识别。轮廓可以简单认为成将连续的点（连着边界）连在一起的曲线，具有相同的颜色或者灰度。

（1）为了更加准确，要使用二值化图像。

（2）在寻找轮廓之前，要进行阈值化处理或者 Canny 边界检测。

（3）绘制轮廓

（4）轮廓的近似

（5）查找轮廓的不同特征 ①角点数 ②面积

通过检测角点来判定是什么形状，当角点无数多时认为为圆形。

当然此思路不适合做复杂环境下的图像检测容易误判，在识别特殊规则的图形比较准确，还有待改进。如有不足，感谢指出。

import cv2 as cv
import numpy as np
global a
class ShapeAnalysis:
    def __init__(self):
        self.shapes = {'three': 0, 'four': 0, 'five':0, 'six': 0, 'circles': 0}

    def analysis(self, frame):
        h, w, ch = frame.shape
        result = np.zeros((h, w, ch), dtype=np.uint8)
        print("start to detect lines...\n")
        # 二值化图像
        gray = cv.cvtColor(frame, cv.COLOR_BGR2GRAY)
        ret, binary = cv.threshold(gray, 0, 255, cv.THRESH_BINARY_INV | cv.THRESH_OTSU)
        cv.imshow("input image", frame)
        #函数cv.findContours有三个参数，第一个是输入图像，第二个是轮廓检索模式，第三个是轮廓近似方法

        out_binary, contours, hierarchy = cv.findContours(binary, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
         #shape数组为面积范围
        areas = np.zeros(shape=[1, 7000])
        areaj = np.zeros(shape=[1, 7000])
        aread = np.zeros(shape=[1, 7000])
        areaw = np.zeros(shape=[1, 7000])
        areal = np.zeros(shape=[1, 7000])
        areayuan = np.zeros(shape=[1, 7000])
        i = 1
        global a
        for cnt in range(len(contours)):
            # 提取与绘制轮廓
            cv.drawContours(result, contours, cnt, (0, 255, 0), 2)

            # 轮廓逼近
            #epsilon，它是从原始轮廓到近似轮廓的最大距离。它是一个准确度参数
            epsilon = 0.01 * cv.arcLength(contours[cnt], True)
            #角点计算
            approx = cv.approxPolyDP(contours[cnt], epsilon, True)
            # 分析几何形状
            corners = len(approx)
            shape_type = ""

            if corners == 3:
                count = self.shapes['three']
                count = count+1
                self.shapes['three'] = count
                #shape_type = "三角形"
                area_3 = cv.contourArea(contours[cnt])
                areas[0, i] = area_3
            if corners == 4:
                count = self.shapes['four']
                count = count + 1
                self.shapes['four'] = count
                #shape_type = "four形"
                area_4 = cv.contourArea(contours[cnt])
                areaj[0, i] = area_4
            if corners == 5:
                count = self.shapes['five']
                count = count + 1
                self.shapes['five'] = count
                #shape_type = "five形"
                area_5 = cv.contourArea(contours[cnt])
                areaw[0, i] = area_5
            #if corners == 6:
           #     count = self.shapes['six']
            #    count = count + 1
            #    self.shapes['six'] = count
             #   #shape_type = "six形"
            #    area_6 = cv.contourArea(contours[cnt])
            #    areal[0, i] = area_6
            if corners >= 50:
                count = self.shapes['circles']
                count = count + 1
                self.shapes['circles'] = count
                #shape_type = "圆形"
                area_y = cv.contourArea(contours[cnt])
                areayuan[0, i] = area_y
            i = i + 1
        area_S = areas.max()
        area_J = areaj.max()
        area_Y = areayuan.max()
        area_D = aread.max()
        area_W = areaw.max()
        area_L = areal.max()
        shape_types = "T"  #三角形
        shape_typej = "F"  #矩形
        shape_typey = "Y"  #圆形
        shape_typew = "F"  #five形
        shape_typel = "S"  #six形

        if area_S >5000:
			a = shape_types
        if area_J >5000:
			a = shape_typej
        if area_Y >5000:
			a = shape_typey
        if area_W>5000:
			a = shape_typew
        if area_L >5000:
			a = shape_typel
        return self.shapes

if __name__ == "__main__":
	cap = cv.VideoCapture(0)
	success, img = cap.read()
	cv.imshow('frame',img)
	cv.imwrite("/home/car/AI_match/4.jpg", img)
	src = cv.imread("/home/car/AI_match/4.jpg")
	ld = ShapeAnalysis()
	ld.analysis(src)
	print(a)
   # cv.waitKey(0)
  #  cv.destroyAllWindows()

图像处理

图形识别

计算机视觉