示例#1
0
def main(args):

    parser = argparse.ArgumentParser(description='PRM Path Planning Algorithm')
    parser.add_argument('--numSamples',
                        type=int,
                        default=1000,
                        metavar='N',
                        help='Number of sampled points')
    args = parser.parse_args()

    numSamples = args.numSamples

    env = open("environment.txt", "r")
    l1 = env.readline().split(";")

    current = list(map(int, l1[0].split(",")))
    destination = list(map(int, l1[1].split(",")))

    print("Current: {} Destination: {}".format(current, destination))

    print("****Obstacles****")
    allObs = []
    for l in env:
        if (";" in l):
            line = l.strip().split(";")
            topLeft = list(map(int, line[0].split(",")))
            bottomRight = list(map(int, line[1].split(",")))
            obs = Obstacle(topLeft, bottomRight)
            obs.printFullCords()
            allObs.append(obs)

    utils = Utils()
    utils.drawMap(allObs, current, destination)

    prm = PRMController(numSamples, allObs, current, destination)
    # Initial random seed to try
    initialRandomSeed = 0
    prm.runPRM(initialRandomSeed)
示例#2
0
def process():
    #Inicializar imagen
    img = cv.imread('..\cameraControllerPython\Image\map.jpg')
    OrigImag = img.copy()

    # Load image
    image_bgr = img
    # Convert to RGB
    image_rgb = cv.cvtColor(image_bgr, cv.COLOR_BGR2RGB)
    # Rectange values: start x, start y, width, height
    rectangle = (25, 25, img.shape[1] - 50, img.shape[0] - 50)
    # Create initial mask
    mask = np.zeros(image_rgb.shape[:2], np.uint8)
    # Create temporary arrays used by grabCut
    bgdModel = np.zeros((1, 65), np.float64)
    fgdModel = np.zeros((1, 65), np.float64)
    # Run grabCut
    cv.grabCut(
        image_rgb,  # Our image
        mask,  # The Mask
        rectangle,  # Our rectangle
        bgdModel,  # Temporary array for background
        fgdModel,  # Temporary array for background
        5,  # Number of iterations
        cv.GC_INIT_WITH_RECT)  # Initiative using our rectangle
    # Create mask where sure and likely backgrounds set to 0, otherwise 1
    mask_2 = np.where((mask == 2) | (mask == 0), 0, 1).astype('uint8')
    # Multiply image with new mask to subtract background
    image_rgb_nobg = image_rgb * mask_2[:, :, np.newaxis]
    # Show image
    plt.imshow(image_rgb_nobg), plt.axis("off")
    plt.show()
    img = cv.cvtColor(image_rgb_nobg, cv.COLOR_RGB2BGR)

    imgray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
    ret, thresh = cv.threshold(imgray, 5, 255, 0)
    contours, hierarchy = cv.findContours(thresh, cv.RETR_TREE,
                                          cv.CHAIN_APPROX_SIMPLE)
    ctrs = img.copy()
    cv.drawContours(ctrs, contours, -1, (0, 255, 0), 3)

    ###cv.imshow('Images', ctrs)
    #print(contours)
    #print(hierarchy)

    decode_preds = {
        0: 'car',
        1: 'destination',
        2: 'house',
        3: 'origin',
        4: 'tree',
        5: 'truck',
        6: 'windmill'
    }
    model = load_model('drone_vision_vgg16.h5')

    xL = OrigImag.shape[0] / 100
    yL = OrigImag.shape[1] / 100
    current = str(round(153 / xL)) + ", " + str(round(478 / yL))
    destination = str(round(1034 / xL)) + ", " + str(round(184 / yL))
    obstacles = np.array(["empty"])

    contoursNew = np.array([[1, 2, 3]])
    divisions = img.copy()
    i = 0
    #print(len(contours))
    while i < len(contours):
        #print(hierarchy[0][i][3])
        (x, y, w, h) = cv.boundingRect(contours[i])
        #print((x,y,w,h))
        if (hierarchy[0][i][3] == -1 and contours[i].shape[0] > 3 and w > 10
                and h > 10):
            #print(cont)
            print("Number: " + str(i))

            divisions = cv.rectangle(divisions, (x, y), (x + w, y + h),
                                     (255, 0, 0), 2)
            #imagTopredict = OrigImag[round(x+w/2-75):round(x+w/2+75), round(y+h/2-75):round(y+h/2+75), :].copy()
            imagTopredict = OrigImag[y:y + h, x:x + w, :].copy()
            #print(imagTopredict.shape)
            imagTopredict = cv.resize(imagTopredict, (150, 150))
            #imagTopredict = array_to_img(imagTopredict)
            # prepare the image for the VGG model
            imagTopredict = imagTopredict / 255
            imagTopredict = np.array(imagTopredict)[:, :, 0:3]
            # convert the image pixels to a numpy array
            ###imagTopredict = img_to_array(imagTopredict)
            # reshape data for the model
            #        if(imagTopredict.shape[0]<imagTopredict.shape[1]):
            #            extension = imagTopredict.shape[0]
            #        else:
            #            extension = imagTopredict.shape[1]

            #imagTopredict = np.stack([imagTopredict], axis=0)
            imagTopredict = imagTopredict.reshape(-1, 150, 150, 3)
            #imagTopredict = imagTopredict.reshape((1, 150, 150, 3))
            #{'DecodeJpeg:0': imagTopredict}
            # predict the probability across all output classes
            prediction = model.predict(imagTopredict)
            if (contours[i].shape[0] == 4):
                prediction[0][3] = 1
            strPred = decode_preds[np.argmax(prediction)]
            strMsg = strPred + ' (' + str(
                round(prediction[0][np.argmax(prediction)] * 100, 2)) + '%)'
            #print(strMsg)
            divisions = cv.putText(divisions, strMsg, (x - 40, y),
                                   cv.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2)
            print(
                f'{decode_preds[np.argmax(prediction)]} ({round(prediction[0][np.argmax(prediction)] * 100, 2)}%)'
            )
            print(cv.HuMoments(cv.moments(contours[i])).flatten())

            if (strPred == 'origin'):
                current = str(round(
                    (x + w / 2) / xL)) + "," + str(round((y + h / 2) / yL))
            elif (strPred == 'destination'):
                destination = str(round(
                    (x + w / 2) / xL)) + "," + str(round((y + h / 2) / yL))
            else:
                temp = str(round(x / xL)) + "," + str(round(
                    y / yL)) + ";" + str(round(
                        (x + w) / xL)) + "," + str(round((y + h) / yL))
                obstacles = np.concatenate((obstacles, [temp]), axis=0)
                contoursNew = np.concatenate(
                    (contoursNew,
                     [[x + w / 2, y + h / 2,
                       math.sqrt(w**2 + h**2) / 2]]),
                    axis=0)
                divisions = cv.circle(divisions,
                                      (round(x + w / 2), round(y + h / 2)),
                                      int(math.sqrt(w**2 + h**2) / 2),
                                      (255, 0, 0), 2)
            i = i + 1
        else:
            contours.pop(i)
            hierarchy = np.delete(hierarchy, i, 1)

    ctrs = img.copy()
    cv.drawContours(ctrs, contours, -1, (0, 255, 0), 3)
    #print
    ###cv.imshow('Images 2', ctrs)
    cv.imshow('Divisions', divisions)
    cv.imwrite("images\divisions.png", divisions)
    '''
    Probabilistic Road Map
    '''

    parser = argparse.ArgumentParser(description='PRM Path Planning Algorithm')
    parser.add_argument('--numSamples',
                        type=int,
                        default=500,
                        metavar='N',
                        help='Number of sampled points')
    args = parser.parse_args()

    numSamples = args.numSamples

    current = list(map(int, current.split(",")))
    destination = list(map(int, destination.split(",")))

    print("Current: {} Destination: {}".format(current, destination))

    print("****Obstacles****")
    allObs = []
    for l in obstacles:
        if (";" in l):
            line = l.strip().split(";")
            topLeft = list(map(int, line[0].split(",")))
            bottomRight = list(map(int, line[1].split(",")))
            obs = Obstacle(topLeft, bottomRight)
            obs.printFullCords()
            allObs.append(obs)

    utils = Utils()
    utils.drawMap(allObs, current, destination)

    numSamples = 250
    prm = PRMController(numSamples, allObs, current, destination)
    # Initial random seed to try
    initialRandomSeed = 0
    xSol, ySol = prm.runPRM(initialRandomSeed)
    xSol = xL * np.array(xSol)
    ySol = yL * np.array(ySol)

    #Print Solution maps
    solutionImg = divisions.copy()
    j = 0
    for j in range(len(ySol) - 1):
        solutionImg = cv.line(solutionImg, (int(xSol[j]), int(ySol[j])),
                              (int(xSol[j + 1]), int(ySol[j + 1])),
                              (255, 0, 0), 2)

    ###cv.imwrite("images\solution.png", solutionImg)

    solutionImgO = OrigImag.copy()
    j = 0
    for j in range(len(ySol) - 1):
        solutionImgO = cv.line(solutionImgO, (int(xSol[j]), int(ySol[j])),
                               (int(xSol[j + 1]), int(ySol[j + 1])),
                               (255, 0, 0), 2)

    ###cv.imwrite("images\solutionMap.png", solutionImgO)

    #xSol = -0.064637985309549*(xSol-157)
    #ySol = -0.065317919075145*(ySol-539)
    contoursNew = np.delete(contoursNew, 0, axis=0)
    print("contoursNew")
    print(contoursNew)
    i = 0
    while i < len(ySol):
        j = i + 2
        iteration = 0
        while j < len(ySol):
            mI = ySol[i]
            b = xSol[i]
            k = (ySol[j] - ySol[i]) / (xSol[j] - xSol[i])
            freeObst = True
            for cN in range(len(contoursNew)):
                x = (contoursNew[cN, 0] + contoursNew[cN, 1] * k - mI * k +
                     b * k**2) / (1 + k**2)
                y = mI + (x - b) * k
                distObs = math.sqrt((contoursNew[cN, 0] - x)**2 +
                                    (contoursNew[cN, 1] - y)**2) - 2
                interX = (x < xSol[j] and x > xSol[i]) or (x > xSol[j]
                                                           and x < xSol[i])
                interY = (y < ySol[j] and y > ySol[i]) or (y > ySol[j]
                                                           and y < ySol[i])
                if (contoursNew[cN, 2] > distObs and interX and interY):
                    freeObst = False
            if (freeObst):
                xSol = np.delete(xSol, range(i + 1, j), axis=0)
                ySol = np.delete(ySol, range(i + 1, j), axis=0)
                j = i + 2
            else:
                j = j + 1
            iteration = iteration + 1
        i = i + 1

    j = 0
    for j in range(len(ySol) - 1):
        solutionImg = cv.line(solutionImg, (int(xSol[j]), int(ySol[j])),
                              (int(xSol[j + 1]), int(ySol[j + 1])),
                              (0, 255, 0), 2)
    cv.imshow('Soution', solutionImg)
    ###cv.imwrite("images\solution.png", solutionImg)

    solutionImgO = OrigImag.copy()
    j = 0
    for j in range(len(ySol) - 1):
        solutionImgO = cv.line(solutionImgO, (int(xSol[j]), int(ySol[j])),
                               (int(xSol[j + 1]), int(ySol[j + 1])),
                               (0, 255, 0), 2)
    cv.imshow('Soution Map', solutionImgO)

    Sol = np.concatenate(([xSol], [ySol], np.ones((1, np.size(xSol, axis=0)))),
                         axis=0)
    #Sol = np.array([[-0.56648, 0.022008, -2.96859],[0.362136, 0.932125,-559.271],[0,0,1]])*Sol
    Sol = np.matmul(
        np.array([[-0.064717, -0.000218, 10.2782],
                  [0.000218, -0.064717, 34.8483], [0, 0, 1]]), Sol)

    cv.waitKey(0)
    return Sol[0, :], Sol[1, :]