def main(args):
code = wrnchAI.license_check_string(args.license_key) if args.license_key \
else wrnchAI.license_check()
if code != 0:
raise RuntimeError(wrnchAI.returncode_describe(code))
params = wrnchAI.PoseParams()
params.bone_sensitivity = wrnchAI.Sensitivity.high
params.joint_sensitivity = wrnchAI.Sensitivity.high
params.enable_tracking = True
# Default Model resolution
params.preferred_net_width = 328
params.preferred_net_height = 184
output_format = wrnchAI.JointDefinitionRegistry.get('j23')
print('Initializing networks...')
estimator = wrnchAI.PoseEstimator(models_path=args.models_dir,
license_string=args.license_key,
params=params,
gpu_id=0,
output_format=output_format)
print('Initialization done!')
options = wrnchAI.PoseEstimatorOptions()
print('Opening webcam...')
with videocapture_context(args.webcam_index) as cap:
visualizer = Visualizer()
joint_definition = estimator.human_2d_output_format()
bone_pairs = joint_definition.bone_pairs()
while True:
_, frame = cap.read()
if frame is not None:
estimator.process_frame(frame, options) ### passing in frame
humans2d = estimator.humans_2d()
### visualizer is using opencv: using frame and draws over it
visualizer.draw_image(frame) ### overlaying skeleton
for human in humans2d:
joints = human.joints()
visualizer.draw_points(joints)
visualizer.draw_lines(joints, bone_pairs)
visualizer.show()
key = cv2.waitKey(1)
if key & 255 == 27:
break
def wrnch_func(args):
print("Inside wrnch_func...")
print(args.pose_estimator, " = selected pose estimator")
print(args.license_key, " = license key")
reset_camera()
input_handler()
# from /usr/src/wrnchAI/wrSamples/python/pose3d_sample.py
code = wrnchAI.license_check_string(args.license_key) if args.license_key \
else wrnchAI.license_check()
if code != 0:
raise RuntimeError(wrnchAI.returncode_describe(code))
# params
params = wrnchAI.PoseParams()
params.bone_sensitivity = wrnchAI.Sensitivity.high
params.joint_sensitivity = wrnchAI.Sensitivity.high
params.enable_tracking = True
# Default Model resolution
params.preferred_net_width = 328
params.preferred_net_height = 184
# model
output_format = wrnchAI.JointDefinitionRegistry.get('j23')
try:
print('Initializing PoseEstimator...')
estimator = wrnchAI.PoseEstimator(models_path=args.models_dir,
license_string=args.license_key,
params=params,
gpu_id=0,
output_format=output_format)
# which type to process frame
options = wrnchAI.PoseEstimatorOptions()
options.estimate_3d = True
print('Initialization done!')
except:
print_exc()
print('Opening webcam...')
with cv2.VideoCapture(-1) as cap:
visualizer = Visualizer()
joint_definition = estimator.human_3d_output_format()
# bone_pairs = joint_definition.bone_pairs()
while True:
# capture frame by frame
_, frame = cap.read()
if frame is not None:
# passing in frame
estimator.process_frame(frame, options)
# get the pose3d's last estimation by PoseEstimator, returns list of Pose2d <<<
humans3d = estimator.humans_3d()
# visualizer is using opencv: using frame and draws over it
visualizer.draw_image(frame) ### overlaying skeleton
for human in humans3d:
joints = human.joints()
visualizer.draw_points(joints)
visualizer.draw_lines(joints)
visualizer.show()
key = cv2.waitKey(1)
if key & 255 == 27:
break
import sys
import cv2
import wrnchAI
from pynput.keyboard import Key, Controller
num_args = len(sys.argv)
if num_args < 2 or num_args > 3:
sys.exit("Usage: python pos")
if num_args == 3:
webcam_index = int(sys.argv[2])
else:
webcam_index = 0
params = wrnchAI.PoseParams()
params.bone_sensitivity = wrnchAI.Sensitivity.high
params.joint_sensitivity = wrnchAI.Sensitivity.high
params.enable_tracking = True
# Default Model resolution
params.preferred_net_width = 328
params.preferred_net_height = 184
output_format = wrnchAI.JointDefinitionRegistry.get('j25')
print('Initializing networks...')
estimator = wrnchAI.PoseEstimator(models_path=sys.argv[1],
params=params,
gpu_id=0,
output_format=output_format)
def main():
ser1 = serial.Serial('COM9', 9600)
code = wrnchAI.license_check_string(licenseKey)
time.sleep(4)
ser1.write('0'.encode())
# setting tracking parameters
params = wrnchAI.PoseParams()
params.bone_sensitivity = wrnchAI.Sensitivity.high
params.joint_sensitivity = wrnchAI.Sensitivity.high
params.enable_tracking = True
# create resolution
params.preferred_net_width = 500
params.preferred_net_height = 500
# format output
output = wrnchAI.JointDefinitionRegistry.get('j23')
print('Initializing networks')
estimate = wrnchAI.PoseEstimator(models_path='..\\..\\..\\..\\bin\\wrModels', license_string=licenseKey,
params=params, gpu_id=0, output_format=output)
print('Initialization done!')
options = wrnchAI.PoseEstimatorOptions()
print('Turning on webcam now')
# run-time vars:
pose_img_path = r"C:\Users\tayse\OneDrive\Desktop\wrnchAI-engine-GPU-1.15.0-Windows-amd64\src\wrnchAI\wrSamples\python\Assets\pic.jpg"
# extract McWICS Letters:
#pose_imgs = np.load(pose_img_path)
#pose_imgs_list = [pose_imgs[i].tolist() for i in range(6)]
pose = vp.request_joint_data(pose_img_path)
with videocapture_context(0) as cap:
visual = Visualizer()
joint_def = estimate.human_2d_output_format()
bone_pairs = joint_def.bone_pairs()
stagger = 0
while True:
#counter_letter = 0
_, frame = cap.read()
if frame is not None:
estimate.process_frame(frame, options)
humans2d = estimate.humans_2d()
visual.draw_image(frame)
for human in humans2d:
visual.draw_points(human.joints())
visual.draw_lines(human.joints(), bone_pairs)
if stagger % 15 == 0:
# gives the coordinates to calculate similarity
input_data = human.joints().tolist() # list of floats
comp_val = vp.compare_dance(input_data, pose) # compare results
if 0 <= comp_val <= 0.08:
# green
ser1.write('3'.encode())
print('excellent')
print(comp_val)
print("GET READY FOR NEXT LETTER!")
elif 0.08 < comp_val <= 0.14:
# green
ser1.write('2'.encode())
print('good')
print(comp_val)
elif 0.14 < comp_val:
# red
ser1.write('1'.encode())
print('bad')
print(comp_val)
if stagger is 15:
stagger = 0
stagger += 1
visual.show()
key = cv2.waitKey(1)
if key & 255 == 27:
break
def handler(event, context):
wrnch_key = getWrchKey()
models_path = "/usr/bin/wrModels"
# todo: detect realsense index
# camera_indexes = checkValidCameraIndex() # list of camera indexes
camera_indexes = checkValidCameraIndex(cameraIndex())
print(camera_indexes)
camera_index = int(camera_indexes[1])
# todo input from event
user = "******"
test_time = time.strftime('%l.%M%p_%b_%d_%Y')
# test time has a leading space
test_time = str(test_time)[1:]
output_path = "/gg/data/{}/{}/".format(user, test_time)
frames_number = 200
testTimeLength = 5
test_type = 'chair'
serial_number = 123
topic = 'joints/{}'.format(str(serial_number))
# todo: fix video save, see line 156
# Define the codec and create VideoWriter object
# fourcc = cv2.VideoWriter_fourcc('h', '2', '6', '4')
# fourcc = cv2.VideoWriter_fourcc(*'X264')
fourcc = cv2.VideoWriter_fourcc(*'XVID')
fps = 20.0
resolution = (640, 480)
video_out = cv2.VideoWriter(output_path + 'output.avi', fourcc, fps,
resolution)
joints_filename = test_time + "_joints.csv"
confidence_filename = test_time + "_confidence_joints.csv"
confidence_human_filename = test_time + "_confidence_human.csv"
# dictionary that will be saved as a json of the payloads
big_payload = {}
# payload to indicate starting
payload = {
'start_test': True,
'test_time': test_time,
'user': user,
'output_path': output_path,
'length_of_test': testTimeLength,
'test_type': test_type,
'serial_number': serial_number,
}
publish_mqtt(json.dumps(payload), topic)
# generate the empty csv files
generateCsv(output_path, joints_filename, confidence_filename,
confidence_human_filename)
"""
This part takes a long time to 'warm up'
It still seems to be checking the license. If used on too many devices, it will still give an error
"""
### SETUP WRNCH
# This appears to not be needed, but will read the license from the environment variable
# if the key is not passed to it.
# code = wrnchAI.license_check_string(wrnch_key) if wrnch_key \
# else wrnchAI.license_check()
# if code != 0:
# raise RuntimeError(wrnchAI.returncode_describe(code))
params = wrnchAI.PoseParams()
params.bone_sensitivity = wrnchAI.Sensitivity.high
params.joint_sensitivity = wrnchAI.Sensitivity.high
params.enable_tracking = True
# Default Model resolution
params.preferred_net_width = 328
params.preferred_net_height = 184
# the part that takes so long
output_format = wrnchAI.JointDefinitionRegistry.get('j25')
estimator = wrnchAI.PoseEstimator(models_path=models_path,
license_string=wrnch_key,
params=params,
gpu_id=0,
output_format=output_format)
options = wrnchAI.PoseEstimatorOptions()
# todo: find realsense index using realsense sdk
# this was here because video_index was reading a string from an input field,
# Don't change until realsense index is built
video_index = int(camera_index)
with videocapture_context(video_index) as cap:
# wrnch
# visualizer = Visualizer()
joint_definition = estimator.human_2d_output_format()
bone_pairs = joint_definition.bone_pairs()
# time record starts
startTime = time.time()
counter = 0
while True:
_, frame = cap.read()
# todo: remove if not needed
# currently not needed
# # dimensions of frame
# width = frame.shape[1]
# height = frame.shape[0]
# # frame counter
# counter += 1
if frame is not None:
# wrnch
estimator.process_frame(frame, options)
humans2d = estimator.humans_2d()
# visualizer.draw_image(frame)
image_time = time.time()
# todo: check to make sure that len(humans2d)>0, if not, this will break
# try/catch might be needed
try:
human = humans2d[0]
except (RuntimeError, TypeError, NameError):
print("The Length of humans2d[0]:", len(humans2d[0]))
print(RuntimeError, TypeError, NameError)
# for human in humans2d:
joints = human.joints()
confidenceScores = human.scores()
confidence = human.score()
# todo: make this an async function with threading, OR buffer and publish at the end
# buffer is probably better option. Threading is already being used by one of the
# wrnch helpers, which is why asyncio is failing
if len(joints) >= 1:
# publish data to mqtt
# send a dictionary that zips joints_labels and timetamp + joints
joints_dict = dict(
zip(joints_labels(),
[image_time - startTime] + joints.tolist()))
joints_confidence_dict = dict(
zip(joints_labels(), [image_time - startTime] +
confidenceScores.tolist()))
# human_confidence is a single float value
# this is the payload for each joint, some metadata still needs to be added
# be sure to make it match the earlier payload
# This payload should indicate that this ins't the first frame => start_test == false
payload = {
'start_test': False,
'test_time': test_time,
'user': user,
'output_path': output_path,
'length_of_test': testTimeLength,
'test_type': test_type,
'serial_number': serial_number,
'joints': joints_dict,
'joints_confidence': joints_confidence_dict,
'human_confidence': confidence
}
"""
This seems to be slowing down record time
"""
# todo: make async or buffer and publish at the end
# Buffer is the best option right now
# publish payload
# publish_mqtt(json.dumps(payload), topic)
# save payload to to big_payload
big_payload[image_time] = payload
# todo: explore a post-process to convert a dictionary to a csv without pandas
# Pandas is only used to build the csv, another script or lambda could convert
# a json to a csv
#
# write dcata to csv
with open(output_path + joints_filename,
'a') as realTimeCSV:
csvWriter = csv.writer(realTimeCSV)
csvWriter.writerow([image_time - startTime] +
joints.tolist())
with open(output_path + confidence_filename,
'a') as realTimeConfidenceCSV:
csvWriter = csv.writer(realTimeConfidenceCSV)
csvWriter.writerow([image_time - startTime] +
confidenceScores.tolist())
with open(output_path + confidence_human_filename,
'a') as realTimeHumanCSV:
csvWriter = csv.writer(realTimeHumanCSV)
csvWriter.writerow([image_time - startTime] +
[str(confidence)])
# wrnch
# visualizer.draw_points(joints)
# visualizer.draw_lines(joints, bone_pairs)
# wrnch
# visualizer.show()
# todo: Fix the video output
# save the video
# video_out = cv2.VideoWriter(output_path + 'motion_analysis.mp4', fourcc, fps, resolution)
# video_out = cv2.VideoWriter(output_path + 'output.avi', -1, 20.0, (640, 480)) # this one might work - tested in windows
# font = cv2.FONT_HERSHEY_SIMPLEX
# cv2.putText(frame,
# image_time,
# (50, 50),
# font, 1,
# (0, 255, 255),
# 2,
# cv2.)
video_out.write(frame)
if time.time() - startTime >= testTimeLength:
# todo: rewrite make_csv to match payload metadata
# make_csv(big_payload, output_path + "all_" + joints_filename)
with open(output_path + "payload.json", "w") as f:
f.write(json.dumps(big_payload))
# todo: remove plots before deploying
plotJoints(output_path, joints_filename, confidence_filename,
confidence_human_filename)
# todo: direct save all csvs for now
# write_to_s3(bucket_name="motion-analysis", file_name=output_path + joints_filename)
# todo: decompose this buffer by reading the dictionay keys and then publishing each value
# Cannot publish entire payload at once
# publish_mqtt(json.dumps(big_payload), topic)
break