From 7617aca1a839b9e8e7f2f21a27a0fdcaca8fba8c Mon Sep 17 00:00:00 2001
From: Edmond Yoo <hj3yoo@uwaterloo.ca>
Date: Sun, 16 Sep 2018 01:58:28 +0000
Subject: [PATCH] quick update on README
---
opencv_dnn.py | 302 +++++++++++++++++++++++++++++++++++++++++++-------
1 files changed, 260 insertions(+), 42 deletions(-)
diff --git a/opencv_dnn.py b/opencv_dnn.py
index 8746c84..bebbc3c 100644
--- a/opencv_dnn.py
+++ b/opencv_dnn.py
@@ -1,13 +1,90 @@
import cv2
import numpy as np
+import imagehash as ih
import os
import sys
+import math
+import random
+from operator import itemgetter
+
+card_width = 315
+card_height = 440
# Disclaimer: majority of the basic framework in this file is modified from the following tutorial:
# https://www.learnopencv.com/deep-learning-based-object-detection-using-yolov3-with-opencv-python-c/
+# www.pyimagesearch.com/2014/08/25/4-point-opencv-getperspective-transform-example/
+def order_points(pts):
+ # initialzie a list of coordinates that will be ordered
+ # such that the first entry in the list is the top-left,
+ # the second entry is the top-right, the third is the
+ # bottom-right, and the fourth is the bottom-left
+ rect = np.zeros((4, 2), dtype="float32")
+
+ # the top-left point will have the smallest sum, whereas
+ # the bottom-right point will have the largest sum
+ s = pts.sum(axis=1)
+ rect[0] = pts[np.argmin(s)]
+ rect[2] = pts[np.argmax(s)]
+
+ # now, compute the difference between the points, the
+ # top-right point will have the smallest difference,
+ # whereas the bottom-left will have the largest difference
+ diff = np.diff(pts, axis=1)
+ rect[1] = pts[np.argmin(diff)]
+ rect[3] = pts[np.argmax(diff)]
+
+ # return the ordered coordinates
+ return rect
+
+
+def four_point_transform(image, pts):
+ # obtain a consistent order of the points and unpack them
+ # individually
+ rect = order_points(pts)
+ (tl, tr, br, bl) = rect
+
+ # compute the width of the new image, which will be the
+ # maximum distance between bottom-right and bottom-left
+ # x-coordiates or the top-right and top-left x-coordinates
+ widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))
+ widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))
+ maxWidth = max(int(widthA), int(widthB))
+
+ # compute the height of the new image, which will be the
+ # maximum distance between the top-right and bottom-right
+ # y-coordinates or the top-left and bottom-left y-coordinates
+ heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))
+ heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))
+ maxHeight = max(int(heightA), int(heightB))
+
+ # now that we have the dimensions of the new image, construct
+ # the set of destination points to obtain a "birds eye view",
+ # (i.e. top-down view) of the image, again specifying points
+ # in the top-left, top-right, bottom-right, and bottom-left
+ # order
+ dst = np.array([
+ [0, 0],
+ [maxWidth - 1, 0],
+ [maxWidth - 1, maxHeight - 1],
+ [0, maxHeight - 1]], dtype="float32")
+
+ # compute the perspective transform matrix and then apply it
+ M = cv2.getPerspectiveTransform(rect, dst)
+ warped = cv2.warpPerspective(image, M, (maxWidth, maxHeight))
+
+ # If the image is horizontally long, rotate it by 90
+ if maxWidth > maxHeight:
+ center = (maxHeight / 2, maxHeight / 2)
+ M_rot = cv2.getRotationMatrix2D(center, 270, 1.0)
+ warped = cv2.warpAffine(warped, M_rot, (maxHeight, maxWidth))
+
+ # return the warped image
+ return warped
+
+
# Get the names of the output layers
def get_outputs_names(net):
# Get the names of all the layers in the network
@@ -17,10 +94,11 @@
# Remove the bounding boxes with low confidence using non-maxima suppression
-def postprocess(frame, outs, classes, thresh_conf, thresh_nms):
+def post_process(frame, outs, thresh_conf, thresh_nms):
frame_height = frame.shape[0]
frame_width = frame.shape[1]
+
# Scan through all the bounding boxes output from the network and keep only the
# ones with high confidence scores. Assign the box's class label as the class with the highest score.
class_ids = []
@@ -42,17 +120,11 @@
confidences.append(float(confidence))
boxes.append([left, top, width, height])
- # Perform non maximum suppression to eliminate redundant overlapping boxes with
- # lower confidences.
- indices = cv2.dnn.NMSBoxes(boxes, confidences, thresh_conf, thresh_nms)
- for i in indices:
- i = i[0]
- box = boxes[i]
- left = box[0]
- top = box[1]
- width = box[2]
- height = box[3]
- draw_pred(frame, class_ids[i], classes, confidences[i], left, top, left + width, top + height)
+ # Perform non maximum suppression to eliminate redundant overlapping boxes with lower confidences.
+ indices = [ind[0] for ind in cv2.dnn.NMSBoxes(boxes, confidences, thresh_conf, thresh_nms)]
+
+ ret = [[class_ids[i], confidences[i], boxes[i]] for i in indices]
+ return ret
# Draw the predicted bounding box
@@ -73,7 +145,96 @@
cv2.putText(frame, label, (left, top), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255))
-def detect_frame(net, classes, img, thresh_conf=0.5, thresh_nms=0.4, in_dim=(416, 416), out_path=None):
+def remove_glare(img):
+ """
+ Inspired from:
+ http://www.amphident.de/en/blog/preprocessing-for-automatic-pattern-identification-in-wildlife-removing-glare.html
+ The idea is to find area that has low saturation but high value, which is what a glare usually look like.
+ """
+ img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
+ _, s, v = cv2.split(img_hsv)
+ non_sat = (s < 32) * 255 # Find all pixels that are not very saturated
+
+ # Slightly decrease the area of the non-satuared pixels by a erosion operation.
+ disk = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
+ non_sat = cv2.erode(non_sat.astype(np.uint8), disk)
+
+ # Set all brightness values, where the pixels are still saturated to 0.
+ v[non_sat == 0] = 0
+ # filter out very bright pixels.
+ glare = (v > 200) * 255
+
+ # Slightly increase the area for each pixel
+ glare = cv2.dilate(glare.astype(np.uint8), disk)
+ glare_reduced = np.ones((img.shape[0], img.shape[1], 3), dtype=np.uint8) * 200
+ glare = cv2.cvtColor(glare, cv2.COLOR_GRAY2BGR)
+ corrected = np.where(glare, glare_reduced, img)
+ return corrected
+
+
+def find_card(img, thresh_c=5, kernel_size=(3, 3), size_ratio=0.3):
+ # Typical pre-processing - grayscale, blurring, thresholding
+ img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+ img_blur = cv2.medianBlur(img_gray, 5)
+ img_thresh = cv2.adaptiveThreshold(img_blur, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY_INV, 5, thresh_c)
+
+ # Dilute the image, then erode them to remove minor noises
+ kernel = np.ones(kernel_size, np.uint8)
+ img_dilate = cv2.dilate(img_thresh, kernel, iterations=1)
+ img_erode = cv2.erode(img_dilate, kernel, iterations=1)
+
+ # Find the contour
+ #img_contour = img_erode.copy()
+ _, cnts, hier = cv2.findContours(img_erode, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
+ if len(cnts) == 0:
+ print('no contours')
+ return []
+ #img_contour = cv2.cvtColor(img_contour, cv2.COLOR_GRAY2BGR)
+
+ # For each contours detected, check if they are large enough and are rectangle
+ cnts_rect = []
+ ind_sort = sorted(range(len(cnts)), key=lambda i: cv2.contourArea(cnts[i]), reverse=True)
+ for i in range(len(cnts)):
+ size = cv2.contourArea(cnts[ind_sort[i]])
+ peri = cv2.arcLength(cnts[ind_sort[i]], True)
+ approx = cv2.approxPolyDP(cnts[ind_sort[i]], 0.04 * peri, True)
+ if size > img.shape[0] * img.shape[1] * size_ratio and len(approx) == 4:
+ cnts_rect.append(approx)
+
+ return cnts_rect
+
+ '''
+ #card_dim = [630, 880]
+ #for cnt in cnts_rect:
+ # pts = np.float32([p[0] for p in cnt])
+ # img_warp = four_point_transform(img, pts)
+
+ # Check which side is longer
+ len_1 = math.sqrt((cnt[0][0][0] - cnt[1][0][0]) ** 2 + (cnt[0][0][1] - cnt[1][0][1]) ** 2)
+ len_2 = math.sqrt((cnt[0][0][0] - cnt[-1][0][0]) ** 2 + (cnt[0][0][1] - cnt[-1][0][1]) ** 2)
+ #print(len_1, len_2)
+
+ orig_corner = np.array([p[0] for p in cnt], dtype=np.float32)
+ if len_1 > len_2:
+ new_corner = np.array([[0, 0], [0, card_dim[1]], [card_dim[0], card_dim[1]], [card_dim[0], 0]], dtype=np.float32)
+ else:
+ new_corner = np.array([[0, 0], [card_dim[0], 0], [card_dim[0], card_dim[1]], [0, card_dim[1]]],
+ dtype=np.float32)
+
+ M = cv2.getPerspectiveTransform(orig_corner, new_corner)
+ img_warp = cv2.warpPerspective(img, M, (card_dim[0], card_dim[1]))
+
+ #cv2.imshow('warp', img_warp)
+ #cv2.waitKey(0)
+ #img_contour = cv2.drawContours(img_contour, cnts_rect, -1, (0, 255, 0), 3)
+ #img_thresh = cv2.cvtColor(img_thresh, cv2.COLOR_GRAY2BGR)
+ #img_erode = cv2.cvtColor(img_erode, cv2.COLOR_GRAY2BGR)
+ #img_dilate = cv2.cvtColor(img_dilate, cv2.COLOR_GRAY2BGR)
+ #return img_thresh, img_erode, img_contour
+ '''
+
+def detect_frame(net, classes, img, thresh_conf=0.5, thresh_nms=0.4, in_dim=(416, 416), display=True, out_path=None):
+ img_copy = img.copy()
# Create a 4D blob from a frame.
blob = cv2.dnn.blobFromImage(img, 1 / 255, in_dim, [0, 0, 0], 1, crop=False)
@@ -84,7 +245,11 @@
outs = net.forward(get_outputs_names(net))
# Remove the bounding boxes with low confidence
- postprocess(img, outs, classes, thresh_conf, thresh_nms)
+ obj_list = post_process(img, outs, thresh_conf, thresh_nms)
+ for obj in obj_list:
+ class_id, confidence, box = obj
+ left, top, width, height = box
+ draw_pred(img, class_id, classes, confidence, left, top, left + width, top + height)
# Put efficiency information. The function getPerfProfile returns the
# overall time for inference(t) and the timings for each of the layers(in layersTimes)
@@ -94,13 +259,32 @@
if out_path is not None:
cv2.imwrite(out_path, img.astype(np.uint8))
+ if display:
+ #no_glare = remove_glare(img_copy)
+ #img_concat = np.concatenate((img, no_glare), axis=1)
+ cv2.imshow('result', img)
+ '''
+ for i in range(len(obj_list)):
+ class_id, confidence, box = obj_list[i]
+ left, top, width, height = box
+ img_snip = img_copy[max(0, top):min(img.shape[0], top + height),
+ max(0, left):min(img.shape[1], left + width)]
+ img_thresh, img_dilate, img_canny, img_hough = find_card(img_snip)
+ img_concat = np.concatenate((img_snip, img_thresh, img_dilate, img_canny, img_hough), axis=1)
+ cv2.imshow('feature#%d' % i, img_concat)
+ '''
+ cv2.waitKey(0)
+ cv2.destroyAllWindows()
+
+ return obj_list
-def detect_video(net, classes, capture, thresh_conf=0.5, thresh_nms=0.4, in_dim=(416, 416), out_path=None):
+def detect_video(net, classes, capture, thresh_conf=0.5, thresh_nms=0.4, in_dim=(416, 416), display=True, out_path=None):
if out_path is not None:
vid_writer = cv2.VideoWriter(out_path, cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 30,
(round(capture.get(cv2.CAP_PROP_FRAME_WIDTH)),
round(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))))
+ max_num_obj = 0
while True:
ret, frame = capture.read()
if not ret:
@@ -108,28 +292,52 @@
print("End of video. Press any key to exit")
cv2.waitKey(0)
break
- '''
- # Create a 4D blob from a frame.
- blob = cv2.dnn.blobFromImage(frame, 1 / 255, in_dim, [0, 0, 0], 1, crop=False)
+ img = frame.copy()
+ obj_list = detect_frame(net, classes, frame, thresh_conf=thresh_conf, thresh_nms=thresh_nms, in_dim=in_dim,
+ display=False, out_path=None)
+ #cnts_rect = find_card(img)
+ max_num_obj = max(max_num_obj, len(obj_list))
+ if display:
+ img_result = frame.copy()
+ #img_result = cv2.drawContours(img_result, cnts_rect, -1, (0, 255, 0), 2)
+ #for i in range(len(cnts_rect)):
+ # pts = np.float32([p[0] for p in cnts_rect[i]])
+ # img_warp = four_point_transform(img, pts)
+ # cv2.imshow('card#%d' % i, img_warp)
+ #for i in range(len(cnts_rect), max_num_obj):
+ # cv2.imshow('card#%d' % i, np.zeros((1, 1), dtype=np.uint8))
+ #no_glare = remove_glare(img)
+ #img_thresh, img_erode, img_contour = find_card(no_glare)
+ #img_concat = np.concatenate((no_glare, img_contour), axis=1)
- # Sets the input to the network
- net.setInput(blob)
+ for i in range(len(obj_list)):
+ class_id, confidence, box = obj_list[i]
+ left, top, width, height = box
+ offset_ratio = 0.1
+ x1 = max(0, int(left - offset_ratio * width))
+ x2 = min(img.shape[1], int(left + (1 + offset_ratio) * width))
+ y1 = max(0, int(top - offset_ratio * height))
+ y2 = min(img.shape[0], int(top + (1 + offset_ratio) * height))
+ img_snip = img[y1:y2, x1:x2]
+ cnts = find_card(img_snip)
+ if len(cnts) > 0:
+ cnt = cnts[-1]
+ pts = np.float32([p[0] for p in cnt])
+ img_warp = four_point_transform(img_snip, pts)
+ img_warp = cv2.resize(img_warp, (card_width, card_height))
+ #img_thresh, img_dilate, img_contour = find_card(img_snip)
+ #img_concat = np.concatenate((img_snip, img_contour), axis=1)
+ cv2.rectangle(img_warp, (22, 47), (294, 249), (0, 255, 0), 2)
+ cv2.imshow('card#%d' % i, img_warp)
+ else:
+ cv2.imshow('card#%d' % i, np.zeros((1, 1), dtype=np.uint8))
+ for i in range(len(obj_list), max_num_obj):
+ cv2.imshow('card#%d' % i, np.zeros((1, 1), dtype=np.uint8))
+ cv2.imshow('result', img_result)
+ #if len(obj_list) > 0:
+ # cv2.waitKey(0)
- # Runs the forward pass to get output of the output layers
- outs = net.forward(get_outputs_names(net))
- # Remove the bounding boxes with low confidence
- postprocess(frame, outs, classes, thresh_conf, thresh_nms)
-
- # Put efficiency information. The function getPerfProfile returns the
- # overall time for inference(t) and the timings for each of the layers(in layersTimes)
- t, _ = net.getPerfProfile()
- label = 'Inference time: %.2f ms' % (t * 1000.0 / cv2.getTickFrequency())
- cv2.putText(frame, label, (0, 15), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255))
- '''
- detect_frame(net, classes, frame,
- thresh_conf=thresh_conf, thresh_nms=thresh_nms, in_dim=in_dim, out_path=None)
- cv2.imshow('result', frame)
if out_path is not None:
vid_writer.write(frame.astype(np.uint8))
cv2.waitKey(1)
@@ -137,24 +345,33 @@
if out_path is not None:
vid_writer.release()
cv2.destroyAllWindows()
- pass
def main():
# Specify paths for all necessary files
- test_path = '../data/test1.mp4'
+ test_path = os.path.abspath('../data/test4.mp4')
+ #weight_path = 'backup/tiny_yolo_10_39500.weights'
+ #cfg_path = 'cfg/tiny_yolo_10.cfg'
+ #class_path = "data/obj_10.names"
weight_path = 'weights/second_general/tiny_yolo_final.weights'
cfg_path = 'cfg/tiny_yolo.cfg'
- class_path = "data/obj.names"
+ class_path = 'data/obj.names'
out_dir = 'out'
if not os.path.isfile(test_path):
print('The test file %s doesn\'t exist!' % os.path.abspath(test_path))
+ return
if not os.path.isfile(weight_path):
print('The weight file %s doesn\'t exist!' % os.path.abspath(test_path))
+ return
if not os.path.isfile(cfg_path):
print('The config file %s doesn\'t exist!' % os.path.abspath(test_path))
+ return
if not os.path.isfile(class_path):
print('The class file %s doesn\'t exist!' % os.path.abspath(test_path))
+ return
+
+ thresh_conf = 0.01
+ thresh_nms = 0.8
# Setup
# Read class names from text file
@@ -162,8 +379,8 @@
classes = [line.strip() for line in f.readlines()]
# Load up the neural net using the config and weights
net = cv2.dnn.readNetFromDarknet(cfg_path, weight_path)
- #net.setPreferableBackend(cv2.dnn.DNN_BACKEND_OPENCV)
- #net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)
+ net.setPreferableBackend(cv2.dnn.DNN_BACKEND_OPENCV)
+ net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)
# Save the detection result if out_dir is provided
if out_dir is None or out_dir == '':
@@ -172,12 +389,13 @@
out_path = out_dir + '/' + os.path.split(test_path)[1]
# Check if test file is image or video
test_ext = test_path[test_path.find('.') + 1:]
+
if test_ext in ['jpg', 'jpeg', 'bmp', 'png', 'tiff']:
img = cv2.imread(test_path)
- detect_frame(net, classes, img, out_path=out_path)
+ detect_frame(net, classes, img, out_path=out_path, thresh_conf=thresh_conf, thresh_nms=thresh_nms)
else:
- capture = cv2.VideoCapture(test_path)
- detect_video(net, classes, capture, out_path=out_path)
+ capture = cv2.VideoCapture(0)
+ detect_video(net, classes, capture, out_path=out_path, thresh_conf=thresh_conf, thresh_nms=thresh_nms)
capture.release()
pass
--
Gitblit v1.10.0