Use of cv2.FastFeatureDetector

Fast feature corner detector

import numpy as np
import cv2

def fast_corners(img,border_buffer=0):
    '''
    Use the fast corner detector to find (x,y) integer corner locations
    :param img_resized:
    :param border_buffer: If >0, prune away kepoints w/in this distance of the boarder
    :return: key point corner locations: list of int cv2.KeyPoints
    '''
    fast = cv2.FastFeatureDetector_create()
    fast.setNonmaxSuppression(True)
    fast.setThreshold(15)
    kp_non_max = fast.detect(img, None) # List of cv2.KeyPoints
    kp_non_max = (cv2.KeyPoint_convert(kp_non_max)).astype(np.int)  # float np.array of Nx2 (array of x,y points)
    if border_buffer > 0:
        kp_non_max = prune_out_of_bound_kp(kp_non_max,img,border_buffer,border_buffer)
    return kp_non_max #.astype(np.int)

def prune_out_of_bound_kp(kp,img,max_offset=30,min_offset=30):
    '''
    Prune keypoings kp that will go out of bounds if a min or max offset is added to them
    NOTE: min_offsets are most offten negative, so you add them below to test if the result is > 0
    :param kp:          cv2.points in a kp list from say Fast corner detection
    :param img:         image you'll be working on to establish the bounds
    :param max_offset:  max offset that could be applied to any kp
    :param min_offset:  min offset that could be applied to any kp
    :return:            filtered list of keypoints that will not go out of image bounds
    '''
    h,w,c = img.shape
    kp_inbounds = []
    for pt in kp:
        x = pt[0]
        x_min = x - min_offset
        x_max = x + max_offset
        if x_min < 0 or x_max >= w:
            continue
        y = pt[1]
        y_min = y - min_offset
        y_max = y + max_offset
        if y_min < 0 or y_max >= h:
            continue
        kp_inbounds.append(pt)
    return kp_inbounds