|
|
@ -1,31 +1,25 @@ |
|
|
|
#!/usr/bin/python3 |
|
|
|
|
|
|
|
import pickle |
|
|
|
import threading |
|
|
|
import numpy as np |
|
|
|
import os |
|
|
|
import sys |
|
|
|
import tensorflow as tf |
|
|
|
import cv2 |
|
|
|
from distutils.version import StrictVersion |
|
|
|
import socket |
|
|
|
from utils import label_map_util |
|
|
|
|
|
|
|
from utils import visualization_utils as vis_util |
|
|
|
import psutil |
|
|
|
import json |
|
|
|
|
|
|
|
import base64 |
|
|
|
from PIL import Image |
|
|
|
from io import BytesIO |
|
|
|
import psutil |
|
|
|
|
|
|
|
switch = 1 |
|
|
|
|
|
|
|
import io |
|
|
|
import socket |
|
|
|
import struct |
|
|
|
import time |
|
|
|
import pickle |
|
|
|
import zlib |
|
|
|
|
|
|
|
# This is needed since the notebook is stored in the object_detection folder. |
|
|
|
sys.path.append("..") |
|
|
@ -40,9 +34,9 @@ if StrictVersion(tf.__version__) < StrictVersion('1.12.0'): |
|
|
|
|
|
|
|
encode_param = [int(cv2.IMWRITE_JPEG_QUALITY), 90] |
|
|
|
|
|
|
|
MODEL_NAME = 'ssd_mobilenet_v1_coco_2017_11_17' #not even worth trying |
|
|
|
#MODEL_NAME="ssd_inception_v2_coco_11_06_2017" # not bad and fast |
|
|
|
#MODEL_NAME="rfcn_resnet101_coco_11_06_2017" # WORKS BEST BUT takes 4 times longer per image |
|
|
|
#MODEL_NAME = 'ssd_mobilenet_v1_coco_2017_11_17' #not even worth trying |
|
|
|
#MODEL_NAME = "ssd_inception_v2_coco_2017_11_17" # not bad and fast |
|
|
|
MODEL_NAME = "rfcn_resnet101_coco_11_06_2017" # WORKS BEST BUT takes 4 times longer per image |
|
|
|
#MODEL_NAME = "faster_rcnn_resnet101_coco_11_06_2017" # too slow |
|
|
|
MODEL_FILE = MODEL_NAME + '.tar.gz' |
|
|
|
DOWNLOAD_BASE = 'http://download.tensorflow.org/models/object_detection/' |
|
|
@ -60,14 +54,14 @@ with detection_graph.as_default(): |
|
|
|
serialized_graph = fid.read() |
|
|
|
od_graph_def.ParseFromString(serialized_graph) |
|
|
|
tf.import_graph_def(od_graph_def, name='') |
|
|
|
|
|
|
|
|
|
|
|
category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS, use_display_name=True) |
|
|
|
|
|
|
|
def load_image_into_numpy_array(image): |
|
|
|
(im_width, im_height) = image.size |
|
|
|
return np.array(image.getdata()).reshape( |
|
|
|
(im_height, im_width, 3)).astype(np.uint8) |
|
|
|
|
|
|
|
|
|
|
|
# For the sake of simplicity we will use only 2 images: |
|
|
|
# image1.jpg |
|
|
|
# image2.jpg |
|
|
@ -82,11 +76,7 @@ data = {"gpu_temp":"10C","gpu_load":"15%","cpu_temp":"47C","cpu_load":"15%","mem |
|
|
|
|
|
|
|
def get_temps(): |
|
|
|
global data |
|
|
|
temps = psutil.sensors_temperatures() |
|
|
|
data["cpu_temp"] = str(int(temps["dell_smm"][0][1]))+"°C" |
|
|
|
data["cpu_load"] = str(psutil.cpu_percent())+"%" |
|
|
|
data["mem_load"] = str(dict(psutil.virtual_memory()._asdict())["percent"])+"%" |
|
|
|
data["fan_speed"] = str(psutil.sensors_fans()["dell_smm"][0][1])+"RPM" |
|
|
|
|
|
|
|
|
|
|
|
def run_inference_for_single_image(image, graph): |
|
|
|
global switch |
|
|
@ -135,82 +125,141 @@ def run_inference_for_single_image(image, graph): |
|
|
|
if 'detection_masks' in output_dict: |
|
|
|
output_dict['detection_masks'] = output_dict['detection_masks'][0] |
|
|
|
|
|
|
|
|
|
|
|
return output_dict |
|
|
|
cut=[-175,-1,-175,-1] |
|
|
|
cut_send = [0,0,0,0] |
|
|
|
a = 1 |
|
|
|
|
|
|
|
kill = True |
|
|
|
|
|
|
|
def listener(port=8385): |
|
|
|
serversocket = socket.socket(socket.AF_INET, socket.SOCK_STREAM) |
|
|
|
serversocket.bind((socket.gethostname(), port)) |
|
|
|
serversocket.listen(5) |
|
|
|
while kill: |
|
|
|
serversocket.accept() |
|
|
|
|
|
|
|
print('Bye!') |
|
|
|
|
|
|
|
|
|
|
|
cut = [115, 100, 400, 150] |
|
|
|
cut_send = [0, 0, 0, 0] |
|
|
|
img_counter = 0 |
|
|
|
socket_switch = True |
|
|
|
cam = cv2.VideoCapture(0) |
|
|
|
dont_send = False |
|
|
|
#cam = cv2.VideoCapture(0) |
|
|
|
cam = cv2.VideoCapture('amb_1.mp4') |
|
|
|
thread = threading.Thread(target=listener) |
|
|
|
thread.start() |
|
|
|
with detection_graph.as_default(): |
|
|
|
sess = tf.Session() |
|
|
|
switch = 0 |
|
|
|
get_temps() |
|
|
|
amb_center = {'x': (400 + 550)/2, 'y': (115+215)/2} |
|
|
|
a = 1 |
|
|
|
# frame_data = [] |
|
|
|
while 1: |
|
|
|
if(True): |
|
|
|
a += 1 |
|
|
|
ret, image = cam.read() |
|
|
|
if a % 10 != 0: |
|
|
|
continue |
|
|
|
try: # Kavşak |
|
|
|
t1 = time.time() |
|
|
|
image_np = image |
|
|
|
image_np_expanded = np.expand_dims(image_np, axis=0) |
|
|
|
output_dict = run_inference_for_single_image(image_np_expanded, detection_graph) |
|
|
|
height, width, channels = image_np.shape |
|
|
|
# frame_data.append(output_dict) |
|
|
|
|
|
|
|
out_dict = {'detection_boxes': [], 'detection_classes': [], 'detection_scores': []} |
|
|
|
for i in output_dict['detection_classes']: |
|
|
|
cont = False |
|
|
|
if i in [3, 6, 8]: # Car, bus, truck |
|
|
|
index = np.where(output_dict['detection_classes'] == i)[0][0] |
|
|
|
score = output_dict['detection_scores'][index] |
|
|
|
if score > 0.3: |
|
|
|
if not any((output_dict['detection_boxes'][index] == b).all() for b in out_dict['detection_boxes']): |
|
|
|
avg_x = (output_dict['detection_boxes'][index][0] + output_dict['detection_boxes'][index][2])/2 |
|
|
|
avg_y = (output_dict['detection_boxes'][index][1] + output_dict['detection_boxes'][index][3])/2 |
|
|
|
for box in out_dict['detection_boxes']: |
|
|
|
avg_box_x = (box[0] + box[2])/2 |
|
|
|
avg_box_y = (box[1] + box[3])/2 |
|
|
|
if abs(avg_x-avg_box_x) < 0.1 and abs(avg_y-avg_box_y) < 0.1: |
|
|
|
cont = True |
|
|
|
continue |
|
|
|
|
|
|
|
if cont: |
|
|
|
continue |
|
|
|
out_dict['detection_classes'].append(i) |
|
|
|
out_dict['detection_boxes'].append(output_dict['detection_boxes'][index]) |
|
|
|
out_dict['detection_scores'].append(output_dict['detection_scores'][index]) |
|
|
|
|
|
|
|
out_dict['detection_classes'] = np.array(out_dict['detection_classes']) |
|
|
|
out_dict['detection_boxes'] = np.array(out_dict['detection_boxes']) |
|
|
|
out_dict['detection_scores'] = np.array(out_dict['detection_scores']) |
|
|
|
for i in out_dict['detection_boxes']: |
|
|
|
(left, right, top, bottom) = (i[1] * width, i[3] * width, |
|
|
|
i[0] * height, i[2] * height) |
|
|
|
if abs(((left + right)/2) - amb_center['x']) < 15 and abs(((top + bottom)/2) - amb_center['y']) < 15: |
|
|
|
print('Ambulance found!') |
|
|
|
|
|
|
|
print(len(out_dict['detection_classes']), ' cars.') |
|
|
|
vis_util.visualize_boxes_and_labels_on_image_array( |
|
|
|
image_np, |
|
|
|
out_dict['detection_boxes'], |
|
|
|
out_dict['detection_classes'], |
|
|
|
out_dict['detection_scores'], |
|
|
|
category_index, |
|
|
|
instance_masks=out_dict.get('detection_masks'), |
|
|
|
use_normalized_coordinates=True, |
|
|
|
line_thickness=8, |
|
|
|
min_score_thresh=0.3 |
|
|
|
) |
|
|
|
cv2.imshow('frame', image_np) |
|
|
|
ex_c = [27, ord("q"), ord("Q")] |
|
|
|
if cv2.waitKey(1) & 0xFF in ex_c: |
|
|
|
break |
|
|
|
|
|
|
|
t2 = time.time() |
|
|
|
print("time taken for {}".format(t2-t1)) |
|
|
|
if dont_send: |
|
|
|
continue |
|
|
|
send_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) |
|
|
|
if socket_switch: |
|
|
|
try: |
|
|
|
client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM) |
|
|
|
client_socket.connect(('192.168.1.36', 8485)) |
|
|
|
connection = client_socket.makefile('wb') |
|
|
|
socket_switch = False |
|
|
|
except: |
|
|
|
socket_switch = True |
|
|
|
continue |
|
|
|
try: |
|
|
|
ret,image = cam.read() |
|
|
|
image_np = image[cut[0]:cut[1],cut[2]:cut[3]] |
|
|
|
#image_np = image_np[int(r[1]):int(r[1]+r[3]),int(r[0]):int(r[0]+r[2])] |
|
|
|
# the array based representation of the image will be used later in order to prepare the |
|
|
|
# result image with boxes and labels on it. |
|
|
|
|
|
|
|
# Expand dimensions since the model expects images to have shape: [1, None, None, 3] |
|
|
|
image_np_expanded = np.expand_dims(image_np, axis=0) |
|
|
|
t1 = time.time() |
|
|
|
# Actual detection. |
|
|
|
output_dict = run_inference_for_single_image(image_np_expanded, detection_graph) |
|
|
|
# Visualization of the results of a detection. |
|
|
|
vis_util.visualize_boxes_and_labels_on_image_array( |
|
|
|
image_np, |
|
|
|
output_dict['detection_boxes'], |
|
|
|
output_dict['detection_classes'], |
|
|
|
output_dict['detection_scores'], |
|
|
|
category_index, |
|
|
|
instance_masks=output_dict.get('detection_masks'), |
|
|
|
use_normalized_coordinates=True, |
|
|
|
line_thickness=8) |
|
|
|
image[cut[0]:cut[1],cut[2]:cut[3]] = image_np |
|
|
|
send_image = cv2.cvtColor(image,cv2.COLOR_BGR2RGB) |
|
|
|
cv2.imshow("Cam",image) |
|
|
|
cv2.imshow("Cut",image_np) |
|
|
|
|
|
|
|
if socket_switch: |
|
|
|
try: |
|
|
|
client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM) |
|
|
|
client_socket.connect(('127.0.0.1', 8485)) |
|
|
|
connection = client_socket.makefile('wb') |
|
|
|
socket_switch = False |
|
|
|
except: |
|
|
|
socket_switch=True |
|
|
|
continue |
|
|
|
try: |
|
|
|
crop_img = send_image.copy(order='C') |
|
|
|
|
|
|
|
crop_img = Image.fromarray(crop_img,"RGB") |
|
|
|
buffered = BytesIO() |
|
|
|
crop_img.save(buffered, format="JPEG") |
|
|
|
img = base64.b64encode(buffered.getvalue()).decode("ascii") |
|
|
|
client_socket.sendall(json.dumps({"image_full":img,"image_sizes":{"x":cut_send[2],"y":cut_send[0],"width":cut_send[3],"height":cut_send[1]},"load":data}).encode('gbk')+b"\n") |
|
|
|
img_counter += 1 |
|
|
|
|
|
|
|
except: |
|
|
|
socket_switch=True |
|
|
|
|
|
|
|
if img_counter % 10 ==0: |
|
|
|
get_temps() |
|
|
|
t2 = time.time() |
|
|
|
print("time taken for {}".format(t2-t1)) |
|
|
|
|
|
|
|
ex_c = [27, ord("q"), ord("Q")] |
|
|
|
if cv2.waitKey(1) & 0xFF in ex_c: |
|
|
|
break |
|
|
|
except KeyboardInterrupt: |
|
|
|
if not socket_switch: |
|
|
|
client_socket.sendall(b"Bye\n") |
|
|
|
cam.release() |
|
|
|
exit(0) |
|
|
|
crop_img = send_image.copy(order='C') |
|
|
|
crop_img = Image.fromarray(crop_img, "RGB") |
|
|
|
buffered = BytesIO() |
|
|
|
crop_img.save(buffered, format="JPEG") |
|
|
|
img = base64.b64encode(buffered.getvalue()).decode("ascii") |
|
|
|
lens = [len(send_image), 0, len(send_image[0])] |
|
|
|
for i in range(0,len(cut), 2): |
|
|
|
if cut[i] < 0: |
|
|
|
cut_send[i] = lens[i] + cut[i] |
|
|
|
cut_send[i+1] = abs(cut[i])-abs(cut[i+1]) |
|
|
|
client_socket.sendall(json.dumps({"image_full":img,"image_sizes":{"x":cut_send[2],"y":cut_send[0],"width":cut_send[3],"height":cut_send[1]},"load":data}).encode('gbk')+b"\n") |
|
|
|
img_counter += 1 |
|
|
|
except: |
|
|
|
socket_switch = True |
|
|
|
|
|
|
|
if img_counter % 10 == 0: |
|
|
|
get_temps() |
|
|
|
except Exception as e: |
|
|
|
print(e) |
|
|
|
break |
|
|
|
|
|
|
|
if not socket_switch: |
|
|
|
client_socket.sendall(b"Bye\n") |
|
|
|
cam.release() |
|
|
|
|
|
|
|
# with open('frame_data.pkl', 'wb') as f: |
|
|
|
# pickle.dump(frame_data, f) |
|
|
|
cv2.destroyAllWindows() |
|
|
|
cam.release() |
|
|
|
kill = False |
|
|
|
thread.join() |