How to detect an arrow having random color in any background?

I am currently working on a task in which I have to detect an arrow. I am very new to openCV, like I have started it just a week ago.

The task 1 is to determine a red color arrow in my live webcam, which I am working on, and this is the SO post that helped me. But, the task 2 is to determine a random arrow in any background, and the linked post doesn't say anything about it. I know this is possible, as that's why they've given it. I'm asking for any code, just some powerful hints?

Here is my current code:

import cv2
import numpy as np

lower = np.array([0, 0, 100])
upper = np.array([50, 50, 255])
cap = cv2.VideoCapture(0)

while True:
    # frame = cv2.imread("redarrow.jpg")
    _, frame = cap.read()
    # frame = cv2.medianBlur(frame, 3)
    # cv2.imshow("bgr", frame)
    mask = cv2.inRange(frame, lower, upper)
    masked = cv2.bitwise_and(frame, frame, mask = mask)
    # cv2.imshow("masked", masked)

    gray = cv2.cvtColor(masked, cv2.COLOR_BGR2GRAY)
    # cv2.imshow("dilated", dilate)
    _, thresh = cv2.threshold(gray, 5, 255, cv2.THRESH_BINARY)

    # kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 1))
    # thresh = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations = 500)
    # kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 1))
    # thresh = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations = 500)

    cv2.erode(thresh, np.ones((1,1)), iterations = 100)
    cv2.dilate(thresh, np.ones((1,1)), iterations = 100)

    thresh = cv2.medianBlur(thresh, 3)
    cv2.imshow("thresh", thresh)

    cntrs, _ = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
    for cntr in cntrs:
        approx = cv2.approxPolyDP(cntr, 0.03 * cv2.arcLength(cntr, True), True)
        pts = [(ary[0][0], ary[0][1]) for ary in approx]
        inc = []
        ang = []
        # print(len(approx), cv2.contourArea(cntr))

        if len(pts) == 7 and cv2.contourArea(cntr) > 1e3:
            for i in range(7):
                if i == 6:
                    theta = np.arctan2(pts[i][1] - pts[0][1], pts[i][0] - pts[0][0])
                else:
                    theta = np.arctan2(pts[i][1] - pts[i+1][1], pts[i][0] - pts[i+1][0])
                inc.append((theta * 180 / np.pi + 180) % 180)
            print(inc)

            for i in range(7):
                if i == 6:
                    diff = abs(inc[i] - inc[0])
                else:
                    diff = abs(inc[i] - inc[i+1])
                ang.append(45 if abs(diff - 90) > abs(diff - 45) else 90)
            if ang.count(45) == 2:
                f = ang.index(45)
                l = ang[::-1].index(45)
                # print(f, l)
                # if l - f == 2:
                    # print(inc[(f+l)//2])
                # print(ang)

    # cv2.imshow("final", thresh)
    # cv2.waitKey(0)
    if cv2.waitKey() == 27: # check for Esc press
        break

cap.release()