OpenMV 线性回归巡线之一：赛道提取

背景介绍：
通过《图像处理基础》能知道视觉就是在图片上找目标。通过《OpenMV 图片对象（目标识别基础）》能知道怎么找目标。

本篇继续这个话题，说说找目标是为了做什么。

提取赛道

线性回归方法的优点是能在视场中的任何位置找到线，缺点是慢。巡线设置为“灰度”、“QQQVGA”来加快速度，由于不跟踪颜色，所以不需要闭自动白平衡。

使用 histeq() 方法提高图像的对比度，以便接下来将图像二值化分离出赛道。

下面用 OpenMV IDE 提取赛道的灰度阀值，如下：

因为 Sugar 从柜子里翻出的老赛道充满了褶皱，所以噪点特别多。下面用 mean(2) 方法消个噪，再看下效果：

现在看上去好多了，接下来把图像通过 binary() 方法二值化，提取出赛道，如下：

赛道有点细，可以通过 erode(1) 让赛道粗一点，如下：

找赛道

1、车灯

小车前放车灯减小环境光对图像的影响，灯选用 WS2812 灯条（8 个灯珠）。WS2812 的驱动代码如下：

# -*- coding: utf-8 -*-import gc
import pybON  = 255
OFF = 0class WS2812:"""Driver for WS2812 RGB LEDs. May be used for controlling single LED or chainof LEDs.Example of use:chain = WS2812(spi_bus=1, led_count=4)data = [(255, 0, 0),    # red(0, 255, 0),    # green(0, 0, 255),    # blue(85, 85, 85),   # white]chain.show(data)Version: 1.0"""buf_bytes = (0x88, 0x8e, 0xe8, 0xee)def __init__(self, spi_bus=1, led_count=1, intensity=1):"""Params:* spi_bus = SPI bus ID (1 or 2)* led_count = count of LEDs* intensity = light intensity (float up to 1)"""self.led_count = led_countself.intensity = intensity# prepare SPI data buffer (4 bytes for each color)self.buf_length = self.led_count * 3 * 4self.buf = bytearray(self.buf_length)# SPI initself.spi = pyb.SPI(spi_bus, pyb.SPI.MASTER, baudrate=3200000, polarity=0, phase=1)# turn LEDs offself.show([])def show(self, data):"""Show RGB data on LEDs. Expected data = [(R, G, B), ...] where R, G and Bare intensities of colors in range from 0 to 255. One RGB tuple for eachLED. Count of tuples may be less than count of connected LEDs."""self.fill_buf(data)self.send_buf()def send_buf(self):"""Send buffer over SPI."""self.spi.send(self.buf)gc.collect()def update_buf(self, data, start=0):"""Fill a part of the buffer with RGB data.Order of colors in buffer is changed from RGB to GRB because WS2812 LEDhas GRB order of colors. Each color is represented by 4 bytes in buffer(1 byte for each 2 bits).Returns the index of the first unfilled LEDNote: If you find this function ugly, it's because speed optimisationsbeated purity of code."""buf = self.bufbuf_bytes = self.buf_bytesintensity = self.intensitymask = 0x03index = start * 12for red, green, blue in data:red = int(red * intensity)green = int(green * intensity)blue = int(blue * intensity)buf[index] = buf_bytes[green >> 6 & mask]buf[index+1] = buf_bytes[green >> 4 & mask]buf[index+2] = buf_bytes[green >> 2 & mask]buf[index+3] = buf_bytes[green & mask]buf[index+4] = buf_bytes[red >> 6 & mask]buf[index+5] = buf_bytes[red >> 4 & mask]buf[index+6] = buf_bytes[red >> 2 & mask]buf[index+7] = buf_bytes[red & mask]buf[index+8] = buf_bytes[blue >> 6 & mask]buf[index+9] = buf_bytes[blue >> 4 & mask]buf[index+10] = buf_bytes[blue >> 2 & mask]buf[index+11] = buf_bytes[blue & mask]index += 12return index // 12def fill_buf(self, data):"""Fill buffer with RGB data.All LEDs after the data are turned off."""end = self.update_buf(data)# turn off the rest of the LEDsbuf = self.bufoff = self.buf_bytes[0]for index in range(end * 12, self.buf_length):buf[index] = offindex += 1def on():data = []for i in range(0,8):data.append((ON, ON, ON))WS2812(spi_bus=2, led_count=8).show(data)def off():data = []for i in range(0,8):data.append((OFF, OFF, OFF))WS2812(spi_bus=2, led_count=8).show(data)

将上述代码放置在 ws2812.py 文件中供其他程序调用。

2、找车道并画线
(1) 首先了解线性回归算法 API 接口 image.get_regression(阀值) 函数：

此函数对图像所有阀值像素进行线性回归计算，返回一个 image.line 对象，可以用 image.draw_line() 画出返回的线条。

对于线性回归使用的最小二乘法这里不需要深究，但需要了解返回值的意义。首先来看两个实验现象：

通过这两个现象，能够总结出 image.get_regression(阀值) 返回值的意义，如下：

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-Ug9vvhL0-1602815269138)(http://weipeng_su.gitee.io/img/19_OpenMV/line_track_12.jpg)]

当我们了解这个规律后，就可以这样算出轨道的方向和角度：

def line_to_theta_and_dir(line):angle_deg = 0direct = 0if line.theta() > 90:angle_deg = 180 - line.theta()direct = 1elif line.theta() < 90:angle_deg = line.theta()if line.theta() == 0:direct = 0else:direct = -1return (direct, angle_deg)

目前找出赛道的完整程序如下：

import sensor, image, time, mathGRAYSCALE_THRESHOLD = (42, 255)
MAG_THRESHOLD = 4def line_to_theta_and_dir(line):angle_deg = 0direct = 0if line.theta() > 90:angle_deg = 180 - line.theta()direct = 1elif line.theta() < 90:angle_deg = line.theta()if line.theta() == 0:direct = 0else:direct = -1return (direct, angle_deg)def line_track(track_color):direct = 0angle_deg = 0line_magnitude = 0img = sensor.snapshot()img.mean(2)img.binary([GRAYSCALE_THRESHOLD])img.crop([0,15,80,60])img.erode(1)line = img.get_regression([track_color], robust=True)if line and (line.magnitude() >= MAG_THRESHOLD):line_magnitude = line.magnitude()if line_magnitude >= MAG_THRESHOLD:line_color = 0if track_color[0] == 0:line_color = 255img.draw_line(line.line(), line_color)direct, angle_deg = line_to_theta_and_dir(line)return direct, angle_deg, line_magnitude

关注作者

欢迎扫码关注我的公众号MultiMCU EDU。

`提示：在公众号“关于我”页面可加作者微信好友。`

`喜欢本文求点赞，有打赏我会更有动力。`