def __init__(self,
config_file="cfg/yolov4-leaky-416.cfg",
weights="yolov4-leaky-416.weights",
data_file='./cfg/coco.data',
darknet_path=os.environ.get('DARKNET_PATH', '../../darknet/'),
thresh=0.4,
queueSize=128,
debug=False):
self.init_done = False
self.stopped = True
self.debug = debug
self.network, self.class_names, self.class_colors = darknet.load_network(
utils.join_strings_as_path([darknet_path, config_file]),
utils.join_strings_as_path([darknet_path, data_file]),
utils.join_strings_as_path([darknet_path, weights]),
batch_size=1)
self.search_box = DARKNET['search_box']
self.width = darknet.network_width(self.network)
self.height = darknet.network_height(self.network)
self.darknet_image = darknet.make_image(self.width, self.height, 3)
self.thresh = thresh
self.Q = object_detection_queue
self.is_night_vision = None
self.org_frame = None
self.resize_frame = None
self.dirs = utils.DirsHandler(DIRS)
self.small_imgs = []
self.first_result = None
self.init_done = True
def __init__(self,
config,
data_file,
weights, batch_size):
self.network, self.class_names, self.class_colors = darknet.load_network(
config,
data_file,
weights,
batch_size=batch_size
)
def load_model():
origin_path = os.getcwd()
path = os.environ['DARKNET_PATH'] + '/'
os.chdir(path) # cd darknet
net, cls, colors = dn.load_network("cfg/yolov4.cfg", "custom/class.data",
"backup/v4/yolov4_best.weights")
print(net, cls)
print("done")
os.chdir(origin_path) # cd ..
return [net, cls, colors]
def __init__(self, id, batch_size = 15, dataset_dir='../others/dds/dataset/WildTrack/src/C'):
# id: string
self.id = id
# video batch size integer
self.batch_size = batch_size
# displacement
self.displacement_check = {}
# Deep SORT encoding. setting is the same for now
self.max_cosine_distance = 0.3
self.nn_budget = None
self.nms_max_overlap = 1.0
self.temp_dir = 'temp-cropped'
os.makedirs(self.temp_dir, exist_ok = True)
self.dataset_dir = dataset_dir + id
# read the total number of file from the server
fnames = sorted(os.listdir(dataset_dir + id))
self.total_frame = len(fnames)
print("Total number of frames: ", str(self.total_frame))
print("Simulating the camera with video frame size 15")
# initiate the yolo v4 network
network, class_names, class_colors = darknet.load_network(
'./darknet/cfg/yolov4.cfg',
'./darknet/cfg/coco.data',
'./darknet/yolov4.weights',
batch_size=1
)
self.network = network
self.class_names = class_names
# initiate the deep sort network
# multi-person tracking
model_filename = 'model_data/mars-small128.pb'
self.encoder = gdet.create_box_encoder(model_filename, batch_size=1)
self.metric = nn_matching.NearestNeighborDistanceMetric("cosine", self.max_cosine_distance, self.nn_budget)
self.tracker = Tracker(self.metric)
print("Camera initiated")
def detect():
results = {}
images = {}
network, class_names, class_colors = darknet.load_network(
CONFIG_FILE_PATH,
DATA_FILE_PATH,
WEIGHT_FILE_PATH,
YOLO_BATCH_SIZE
)
try:
print('Image recognition started!')
while True:
# cv2.waitKey(50)
direction, x_coordinate , y_coordinate = 0, 0, 0
frame = retrieveImg()
image, detections = imageDetection(frame, network, class_names, class_colors, THRESH)
#structure: list
#element structure: (id, confidence, (bbox))
#bbox: x, y, w, h
for i in detections:
id = i[0]
confidence = i[1]
bbox = i[2]
#Code for evaluating bbox area
print('ID detected: ' + id, ', confidence: ' + confidence)
print('Bounding box: x = ' + str(bbox[0]) + ' y = ' + str(bbox[1]) + ' Width = ' + str(bbox[2]) + ' Height =' + str(bbox[3]))
dist_est(bbox[3], id, direction, x_coordinate, y_coordinate, int(bbox[0]))
if id in symb_confidence:
if id in results:
print('ID has been detected before')
# if float(confidence) > float(results[id][1]):
if float(confidence) > float(results[id][0][1]) and max(symb_confidence[id].keys()) >= max(results[id][1].keys()):
print('Higher confidence. Replacing existing image.')
# Removes existing result from dict
del results[id]
# Removes existing img from dict
del images[id]
# Adds new result to dict
# results[id] = i
# results[id] = [i, symb_confidence[id][max(symb_confidence[id].keys())]]
results[id] = [i, symb_confidence[id]]
# Adds new result to dict
images[id] = image
# saveImage(image, id)
# sendAndroidString(results)
else:
print('Lower confidence. Keeping existing image.')
pass
else:
print('New ID. Saving to results and images dict.')
# results[id] = i
# results[id] = [i, symb_confidence[id][max(symb_confidence[id].keys())]]
results[id] = [i, symb_confidence[id]]
images[id] = image
# saveImage(image, id)
# sendAndroidString(results)
except KeyboardInterrupt:
print('Image recognition ended')
result_string = '{'
print("Results:")
for id in results:
x_coordinate , y_coordinate = results[id][1][max(symb_confidence[id].keys())][0], results[id][1][max(symb_confidence[id].keys())][1]
id_coordinate_str = '(' + id + ',' + str(x_coordinate) + ',' + str(y_coordinate) + '),'
result_string += id_coordinate_str
print('ID: ' + id + ', Coordinates: (' + str(x_coordinate) +',' + str(y_coordinate) + ')' + ', Confidence: ' + results[id][1])
if result_string[-1] == ',':
result_string = result_string[:-1]
result_string += '}'
print(result_string)
# android_full_string = sendAndroidString(results)
# print('Sent ' + android_full_string + ' to Android.')
#generate image mosaic
result_list = list(images.values())
showImages(result_list)
"""
import os
import json
from bedrock_client.bedrock.metrics.service import ModelMonitoringService
from flask import Flask, Response, current_app, request
import cv2
from PIL import Image
import darknet.darknet as darknet
DATA_FOLDER = os.environ.get("DATA_FOLDER")
YOLO_VARS = darknet.load_network(
config_file="darknet/cfg/yolov4-tiny-custom.cfg",
data_file=f"./darknet/{DATA_FOLDER}/obj.data",
weights="/artefact/yolov4-tiny-custom_latest.weights",
batch_size=1,
)
def image_detection(image_path, network, class_names, class_colors, thresh):
"""Run YOLOv4 models."""
# Darknet doesn't accept numpy images.
# Create one with image we reuse for each detect
width = darknet.network_width(network)
height = darknet.network_height(network)
darknet_image = darknet.make_image(width, height, 3)
image = cv2.imread(image_path)
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image_resized = cv2.resize(image_rgb, (width, height),
def detect():
results = {}
images = {}
network, class_names, class_colors = darknet.load_network(
CONFIG_FILE_PATH, DATA_FILE_PATH, WEIGHT_FILE_PATH, YOLO_BATCH_SIZE)
try:
# if (run):
print('Image recognition started!')
prev_dir, prev_x, prev_y = -1, -1, -1
direction, x_coordinate, y_coordinate = -1, -1, -1
while len(results) < 5 or (confidenceBool and checkConfidence()):
# while len(results) < 5:
direction, x_coordinate, y_coordinate = getPosition()
if direction == -1:
break
if direction != prev_dir or x_coordinate != prev_x or y_coordinate != prev_y:
time.sleep(0.05)
frame = retrieveImg()
image, detections = imageDetection(frame, network, class_names,
class_colors, THRESH)
#structure: list
#element structure: (id, confidence, (bbox))
#bbox: x, y, w, h
for i in detections:
id = i[0]
confidence = i[1]
bbox = i[2]
#Prototype: Skip side images using condition on width / height ratio
if float(bbox[2]) / float(bbox[3]) < 0.6:
continue
print('ID detected: ' + id, ', confidence: ' + confidence)
#Code for evaluating bbox area
# print('Bounding box: Width = ' + str(bbox[2]) + ' Height =' + str(bbox[3]))
dist_est(bbox[3], id, direction, x_coordinate,
y_coordinate, round(float(bbox[0]), 1))
if id in symb_confidence:
if id in results:
print('ID has been detected before')
# if float(confidence) > float(results[id][1]):
#if float(confidence) > float(results[id][0][1]) and max(symb_confidence[id].keys()) >= max(results[id][1].keys()):
if float(confidence) > 98.0 and max(
symb_confidence[id].keys()) >= max(
results[id][1].keys()):
print(
'Higher confidence. Replacing existing image.'
)
# Removes existing result from dict
del results[id]
# Removes existing img from dict
del images[id]
# Adds new result to dict
# results[id] = i
# results[id] = [i, symb_confidence[id][max(symb_confidence[id].keys())]]
results[id] = [i, symb_confidence[id]]
# Adds new result to dict
images[id] = image
saveImage(image, id)
sendAndroidString(results)
else:
print(
'Lower confidence. Keeping existing image.'
)
pass
else:
print('New ID. Saving to results and images dict.')
# results[id] = i
# results[id] = [i, symb_confidence[id][max(symb_confidence[id].keys())]]
results[id] = [i, symb_confidence[id]]
images[id] = image
saveImage(image, id)
sendAndroidString(results)
prev_dir, prev_x, prev_y = direction, x_coordinate, y_coordinate
# print(results)
# else:
# raise Exception
except KeyboardInterrupt:
print('Image recognition ended')
result_string = '{'
print("Results:")
for id in results:
image = results[id]
x_coordinate, y_coordinate = image[1][max(
image[1].keys())][0], image[1][max(image[1].keys())][1]
id_coordinate_str = '(' + id + ',' + str(x_coordinate) + ',' + str(
y_coordinate) + '),'
result_string += id_coordinate_str
print('ID: ' + id + ', Coordinates: (' + str(x_coordinate) + ',' +
str(y_coordinate) + ')' + ', Confidence: ' + str(image[0][1]))
if result_string[-1] == ',':
result_string = result_string[:-1]
result_string += '}'
print(result_string)
android_full_string = sendAndroidString(results)
print('IMG - Sent to Android:' + android_full_string)
ir_socket.send("STOP".encode(FORMAT))
# ir_socket.send("AR-T".encode(FORMAT))
showFinalImage(SAVED_IMAGE_PATH)
# result_list = list(images.values())
# showImages(result_list)
return