Skip to content

A AI System Recognition physical object, traffic jams throught sensor, camera, controller by human

Notifications You must be signed in to change notification settings

flightstar/Self-Driving-Car

Repository files navigation

Self Driving Car

AI System Recognition physical object, traffic jams throught sensor, camera, controller by human

Combination of areas as robotics, self-driving cars, motion planning, sensor fusion, Lidar, radar, camera, IMU, GPS, Apollo, embedded system, machine learning, system architecture, connectivity, real-time system

Operation System:

Embedded Linux: References https://www.engineersgarage.com/tutorials/embedded-linux-basics-programming

Embedded System

Autonomous Systems

Realtime communication (RTC)

Robottics: Là nghành robot học nghiên cứu robot có những chức năng như con người (đứa trẻ)

Hardware in loop: Là mô phỏng phát triển và kiểm thử hệ thống real time phức tạp mà không cần quá trình điều khiển và dây chuyền thực tế (bao gồm phần cứng và phần mềm) tức là chuyển dây chuyền xử lý trong phòng thí nghiệm như:

  • Thử nghiệm phần cứng và phần mềm điều khiển ở điều kiện môi trường tới hạn trong phòng thí nghiệm (như nhiệt độ cao/thấp, gia tốc lớn và các cú sốc cơ học, thiết bị kích thích, tính tương thích điện từ).
  • Thử nghiệm các tác động của lỗi và tình trạng không mong đợi của cơ cấu chấp hành, cảm biến và máy tính trên toàn bộ hệ thống.
  • Vận hành và thử nghiệm các điều kiện vận hành tới hạn và nguy hiểm.
  • Các thử nghiệm tái sinh, có thể lặp lại thường xuyên.
  • Vận hành dễ dàng với các giao diện người-máy khác nhau như buồng tập lái mô phỏng máy bay

Bạn nên hiểu bản chất của những phương pháp xử lý ảnh camera để tạo sự thú vị và yêu thích môn học về xe không người lái.

Select colour

Source code

Source Code

Source Code

Source Code

Là thuật toán dò các cạnh được John F. Canny đưa ra vào năm 1986

Mainroad Mainroad

Source Code

Region of Interest Selection

Hough Space Transformation of the Line Detection

In image space, a line is plotted as x vs. y, but in 1962, Paul Hough devised a method for representing lines in parameter space, which we will call Hough space in his honor. Read more about Hough Transform

In Hough space, I can represent my x vs. y line as a point in m vs. b instead. The Hough Transform is just the conversion from image space to Hough space. So, the characterization of a line in image space will be a single point at the position (m, b) in Hough space.

Implementation

Mainroad Mainroad

Source Code

Source Code

Project: Detect Land through Camera

Source Code

Terminology & analysis

các hình ảnh input đầu vào

Color Selection
import numpy as np
import cv2
def select_rgb_white_yellow(image): 
    # white color mask
    lower = np.uint8([200, 200, 200])
    upper = np.uint8([255, 255, 255])
    white_mask = cv2.inRange(image, lower, upper)
    # yellow color mask
    lower = np.uint8([190, 190,   0])
    upper = np.uint8([255, 255, 255])
    yellow_mask = cv2.inRange(image, lower, upper)
    # combine the mask
    mask = cv2.bitwise_or(white_mask, yellow_mask)
    masked = cv2.bitwise_and(image, image, mask = mask)
    return masked
 image = select_rgb_white_yellow("/Resource/test_image.png")
 cv2.imshow('test_image', image)

import numpy as np
import cv2

# HSV Color Space
def convert_hsv(image):
    return cv2.cvtColor(image, cv2.COLOR_RGB2HSV)
image = convert_hsv("/Resource/test_image.png")
cv2.imshow('test_image', image)

import numpy as np
import cv2

# HLS Colour Space
def convert_hls(image):
    return cv2.cvtColor(image, cv2.COLOR_RGB2HLS)
image = convert_hsv("/Resource/test_image.png")
cv2.imshow('test_image', image)

  • cv2.inRange: Filter the white color and the yellow color seperately. The function returns 255 when the filter conditon is satisfied. Otherwise, it returns 0.
  • cv2.bitwise_or to combine these two binary masks. The combined mask returns 255 when either white or yellow color is detected.
  • cv2.bitwise_and to apply the combined mask onto the original RGB image
import os, glob
import numpy as np
import cv2

def select_white_yellow(image):
    converted = convert_hls(image)
    lower = np.uint8([  0, 200,   0])
    upper = np.uint8([255, 255, 255])
    white_mask = cv2.inRange(converted, lower, upper)
    lower = np.uint8([ 10,   0, 100])
    upper = np.uint8([ 40, 255, 255])
    yellow_mask = cv2.inRange(converted, lower, upper)
    mask = cv2.bitwise_or(white_mask, yellow_mask)
    return cv2.bitwise_and(image, image, mask = mask)

test_images = []
test_images = [plt.imread(path) for path in glob.glob('test_images/*.jpg')]
white_yellow_images = list(map(select_white_yellow, test_images))

Canny Edge Detection Algorithm
  • use cv2.cvtColor to convert images into gray scale
  • use cv2.GaussianBlur to smooth out rough edges
  • use cv2.Canny to find edges
Gray Scaling

Converted into gray scaled ones in order to detect shapes (edges) in the images. This is because the Canny edge detection measures the magnitude of pixel intensity changes or gradients. I converted the white and yellow line images from the above into gray scale for edge detection.

def convert_gray_scale(image):
    return cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
gray_images = list(map(convert_gray_scale, white_yellow_images))

Gaussian Smoothing

The above images have many rough edges which causes many noisy edges to be detected. We use cv2.GaussianBlur to smooth out edges.

def apply_smoothing(image, kernel_size=15):
#   kernel_size must be postivie and odd
#   kernel_size value requires more time to process
    return cv2.GaussianBlur(image, (kernel_size, kernel_size), 0)
blurred_images = list(map(lambda image: apply_smoothing(image), gray_images))

  • If a pixel gradient is higher than the upper threshold, the pixel is accepted as an edge
  • If a pixel gradient value is below the lower threshold, then it is rejected.
  • If the pixel gradient is between the two thresholds, then it will be accepted only if it is connected to a pixel that is above the upper threshold.
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import numpy as np
import cv2

def show_images(images):
    cols = 3
    rows = 2
    plt.figure(figsize=(15, 5))
    for i in range(0, len(images)):
        plt.subplot(rows, cols, i+1)
        plt.imshow(images[i], cmap="gray")
    plt.show()
    
def canny(image):
    low_threshold = 50
    high_threshold = 150
    return cv2.Canny(image, low_threshold, high_threshold)
    
if __name__ == "main":
   edge_images = list(map(lambda image: canny(image), blurred_images))
   show_images(edge_images)

Detect the edges
Region of Interest Selection Method

When finding lane lines, we don't need to check the sky and the hills.

def filter_region(image, vertices):
#   Create the mask using the vertices and apply it to the input image
    mask = np.zeros_like(image)
#   Defining a 3 channel, 2 channel or 1 channel color to fill the mask with depending on the input image
    if len(mask.shape) == 2:
        cv2.fillPoly(mask, vertices, 255)
    else:
        cv2.fillPoly(mask, vertices, (255,)*mask.shape[2]) # in case, the input image has a channel dimension        
    return cv2.bitwise_and(image, mask)
    
def select_region(image):
    # first, define the polygon by vertices
    rows, cols = image.shape[:2]
    bottom_left  = [cols*0.1, rows*0.95]
    top_left     = [cols*0.4, rows*0.6]
    bottom_right = [cols*0.9, rows*0.95]
    top_right    = [cols*0.6, rows*0.6] 
    # the vertices are an array of polygons (i.e array of arrays) and the data type must be integer
    vertices = np.array([[bottom_left, top_left, top_right, bottom_right]], dtype=np.int32)
    return filter_region(image, vertices)

roi_images = list(map(select_region, edge_images))
show_images(roi_images)

Hough Transform Line Detection
  • Using cv2.HoughLinesP to detect lines in the edge images

  • Several parameters of the above function:

    • rho: Distance resolution of the accumulator in pixels.
    • theta: Angle resolution of the accumulator in radians.
    • threshold: Accumulator threshold parameter. Only those lines are returned that get enough votes (> threshold).
    • minLineLength: Minimum line length. Line segments shorter than that are rejected.
    • maxLineGap: Maximum allowed gap between points on the same line to link them.
    import matplotlib.pyplot as plt
    import matplotlib.image as mpimg
    import numpy as np
    import cv2
    import os, glob
    
    def draw_lines(img, lines, color = [255, 0, 0], thickness = 2):
      imgCopy = np.copy(img)
      for line in lines:
          for x1,y1,x2,y2 in line:
              cv2.line(imgCopy, (x1, y1), (x2, y2), color, thickness)
      return imgCopy
    
    def hough_lines(img):
      rho = 1
      theta = np.pi/180
      threshold = 20
      min_line_length = 20
      max_line_gap = 300
      return cv2.HoughLinesP(img, rho, theta, threshold, np.array([]), minLineLength=min_line_length, maxLineGap=max_line_gap)
      
    if __name__ == '__main__':
      list_of_lines = list(map(hough_lines, roi_images))
      line_images = []
      # zip() used for mapping similar index, value of multiple containers of zip() compress
      for image, lines in zip(test_images, list_of_lines):
          line_images.append(draw_lines(image, lines))
      show_images(line_images)

Extrapolating Land Lines and Average them

These lines are not smooth in image. Because there are multiple lines in images. We should get them and add an average line instead of them.

We'll collect positive slope lines and negative slope lines separately and take averages.

def average_slope_intercept(lines):
    left_lines    = [] # (slope, intercept)
    left_weights  = [] # (length,)
    right_lines   = [] # (slope, intercept)
    right_weights = [] # (length,)
    
    for line in lines:
        for x1, y1, x2, y2 in line:
            if x2==x1:
                continue # ignore a vertical line
            # Do doc
            slope = (y2-y1)/(x2-x1)
            # Mat phang duong
            intercept = y1 - slope*x1
            # Dien tich cua line bao quanh
            length = np.sqrt((y2-y1)**2+(x2-x1)**2)
            if slope < 0: # y is reversed in image
                left_lines.append((slope, intercept))
                left_weights.append((length))
            else:
                right_lines.append((slope, intercept))
                right_weights.append((length))
    # add more weight to longer lines    
    left_lane  = np.dot(left_weights,  left_lines) /np.sum(left_weights)  if len(left_weights) >0 else None
    right_lane = np.dot(right_weights, right_lines)/np.sum(right_weights) if len(right_weights)>0 else None
    
    return left_lane, right_lane # (slope, intercept), (slope, intercept)

def lane_lines(image, lines):
    left_lane, right_lane = average_slope_intercept(lines)
    
    y1 = image.shape[0] # bottom of the image
    y2 = y1*0.6         # slightly lower than the middle
    
#   Convert a line represented in slope and intercept into pixel points
    if (left_lane is None) or (right_lane is None) :
        raise Exception('not detecting')

    slope, intercept = left_lane
    x1 = int((y1 - intercept)/slope)
    x2 = int((y2 - intercept)/slope)
    y1 = int(y1)
    y2 = int(y2)
    
    left_line = ((x1, y1), (x2, y2))
    slope, intercept = right_lane
    x1 = int((y1 - intercept)/slope)
    x2 = int((y2 - intercept)/slope)
    y1 = int(y1)
    y2 = int(y2)
    
    right_line = ((x1, y1), (x2, y2))
    return left_line, right_line

def draw_lane_lines(image, lines, color=[255, 0, 0], thickness=20):
    line_image = np.zeros_like(image)
    for line in lines:
        if line is not None:
            cv2.line(line_image, *line,  color, thickness)
    return cv2.addWeighted(image, 1.0, line_image, 0.95, 0.0)
             
if __name__ == '__main__':
    lane_images = []
    for image, lines in zip(test_images, list_of_lines):
        lane_images.append(draw_lane_lines(image, lane_lines(image, lines)))    
    show_images(lane_images)

Image Lane Finding Pipeline
def process_image(image):
    try :
        white_yellow_image = select_white_yellow(image)
        gray_image = convert_gray_scale(image)
        blurred_image = apply_smoothing(gray_image)
        edge_image = canny(blurred_image)
        roi_image = select_region(edge_image)
        list_of_lines = hough_lines(roi_image)
        result = draw_lane_lines(image, lane_lines(image, list_of_lines))
        return result
    except Exception as error:
        return image

# Example: Resource/*.jpg
test_images = [plt.imread(path) for path in glob.glob('Resource/*.jpg')]
lane_images = []
for image in test_images :
    lane_images.append(process_image(image))
show_images(lane_images)

About

A AI System Recognition physical object, traffic jams throught sensor, camera, controller by human

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages