Free Motion Capture for Everyone πβ¨
TypeScript
Tiles camera windows across screen
Jon PC Win Hallway012a378f4086
diff --git a/jon_scratch/multi_camera_handler.py b/jon_scratch/multi_camera_handler.pyindex c9037eaf..15d7784e 100644--- a/jon_scratch/multi_camera_handler.py+++ b/jon_scratch/multi_camera_handler.py@@ -1,3 +1,4 @@+#%%import loggingfrom tracemalloc import startfrom typing import List@@ -35,7 +36,7 @@ def stuff_incoming_frames_into_a_queue(camera: OpenCVCamera, thread_queue: queuelogger.info('{} grabed a frame at timestamp {}'.format(camera.name, timestamp_ns/1e9))if not success:logger.error('{} failed to grab a frame at timestamp {}'.format(camera.name, timestamp_ns/1e9))- thread_queue.put((success, camera.name, image, timestamp_ns))+ thread_queue.put((success, camera.port_number, camera.name, image, timestamp_ns))@@ -59,133 +60,193 @@ def display_image_on_its_cv2_named_window(cam_window_name: str, image: np.ndarraif thread_exit_event.is_set():cv2.destroyAllWindows()-def create_video_viewing_window(camera_name:str = 'Cam'):+def create_video_viewing_window(camera_name:str = 'Cam', window_location = (0,0)):window_name = camera_name + " - Press ESC to exit"cv2.namedWindow(window_name)+ cv2.moveWindow(window_name, window_location[0], window_location[1])return window_name-if __name__ == '__main__':-- camera_found_bool = True- camera_list = []- this_port_number = -1- while camera_found_bool:- this_port_number+=1- camera = OpenCVCamera(port_number=this_port_number)- camera_found_bool = camera.connect()- if camera_found_bool:- camera_list.append(camera)-- global thread_exit_event- thread_exit_event = threading.Event()- loop_count = 0- camera_thread_list = [None]*len(camera_list)- incoming_frames_queue = queue.Queue()-- #%% Start the camera threads (THIS IS WHERE THE MAGIC HAPPENS)- for this_camera in camera_list:- camera_thread_list[this_camera.port_number] = start_camera_thread(this_camera, incoming_frames_queue)-- #%% setting up to grab incoming timestamps etc- dict_of_timestamp_lists = {}- dict_of_fps_mean_lists = {}- dict_of_fps_std_lists = {}- for this_camera in camera_list:- dict_of_timestamp_lists[this_camera.name] = []- dict_of_fps_mean_lists[this_camera.name] = []- dict_of_fps_std_lists[this_camera.name] = []-- #create cv2 named windows where we'll display incoming frames- cam_window_name_dict = {}- for this_camera in camera_list:- cam_window_name_dict[this_camera.name] = create_video_viewing_window(this_camera.name)-- # start the MainThread loop- approx_start_time = time.time()- end_after_seconds = 15 #or on pressing ESC- image_scale = 1 #downsize images if the queue starts to fill up- qsize = []- while not thread_exit_event.is_set():- time_elapsed = time.time() - approx_start_time- this_qsize = incoming_frames_queue.qsize()- qsize.append(this_qsize)- print('Time remaining: {:.2f} - Queueue size: {} - Image scale: {}'.format(end_after_seconds - time_elapsed, this_qsize, image_scale))- if time_elapsed > end_after_seconds:- thread_exit_event.set()- cv2.destroyAllWindows()-- if not incoming_frames_queue.empty():- this_frame_tuple = incoming_frames_queue.get() #<- grab the most recent frame_tuple and do stuff with it- this_success_bool, this_camera_name, this_image, this_timestamp_ns = this_frame_tuple-- if this_qsize < 10:- image_height, image_width, image_channels = this_image.shape- scale_factor = 4- scale_height = int(image_height/scale_factor)- scale_width = int(image_width/scale_factor)-- display_image_on_its_cv2_named_window(cam_window_name_dict[this_camera_name], cv2.resize(this_image, (image_width, image_height)))-- this_timestamp_sec = this_timestamp_ns/1e9-- dict_of_timestamp_lists[this_camera_name].append(this_timestamp_sec)-- list_lengths = [len(this_thing) for this_thing in dict_of_timestamp_lists ]- for this_camera in camera_list:- this_cam_timestamps = np.array(dict_of_timestamp_lists[this_camera.name])-- number_of_frames = np.min([len(dict_of_timestamp_lists[this_key]) for this_key in dict_of_timestamp_lists ]) #truncate all timestamp lists to match frame count of shortest- out_timestamp_dict = {}-- for this_cam_num, this_camera in enumerate(camera_list):- out_timestamp_dict[this_camera.name] = np.array(dict_of_timestamp_lists[this_camera.name][:number_of_frames])- out_dataframe = pd.DataFrame(out_timestamp_dict)- out_dataframe.to_csv('~/timestamps.csv')-- #%% Plot up them bad bois- # where 'bad boi' is defined as a timestamp from each camera- # plt.ion() #<- allows for fiddling, but need to set breakpoint and call plt.show() to see the plots- figure = plt.figure(3214, figsize=(15,5))- ax_timestamps = figure.add_subplot(141)- ax_timestamps.set_xlabel('frame number')- ax_timestamps.set_ylabel('timestamp (unix epoch, sec)')-- ax_frame_duration = figure.add_subplot(142)- ax_frame_duration.set_xlabel('frame number')- ax_frame_duration.set_ylabel('framerate (1/np.diff(timestamp, frames/sec)')- ax_frame_duration.set_ylim(0, .1)-- ax_histogram = figure.add_subplot(143)- ax_histogram.set_xlabel('bins (fps)')- ax_histogram.set_ylabel('histogram count(out of {} frames'.format(number_of_frames))-- ax_qsize = figure.add_subplot(144)- ax_qsize.set_xlabel('frame/loop count')- ax_qsize.set_ylabel('incoming frames queueue size')- ax_qsize.plot(qsize)-- for this_camera in camera_list:- this_cam_timestamps = out_timestamp_dict[this_camera.name]- this_cam_frame_duration = np.diff(out_timestamp_dict[this_camera.name])- this_cam_framerate = 1/this_cam_frame_duration-- ax_timestamps.plot(this_cam_timestamps, 'o',markersize=1, alpha=.5, label=this_camera.name)- ax_frame_duration.plot(this_cam_frame_duration, '-o', markersize=1, alpha=.5, label=this_camera.name)- ax_histogram.hist(this_cam_framerate, bins=np.arange(0,50,1), alpha=.5,label=this_camera.name)-- #reference lines (i.e. 'ideal' numbers)- ideal_framerate = 30- approx_start_time = this_cam_timestamps[0] #just for the last camera, but it's fine for this visualization- ideal_end_time = approx_start_time + number_of_frames/ideal_framerate- ideal_timestamps = np.linspace(approx_start_time, ideal_end_time, number_of_frames)-- ideal_line_label = 'ideal for {}fps'.format(ideal_framerate)- ax_frame_duration.hlines(1/ideal_framerate, 0, number_of_frames, linestyles='dashed', label=ideal_line_label)- ax_histogram.vlines(ideal_framerate, 0, number_of_frames, linestyles='dashed', label=ideal_line_label)- ax_timestamps.plot(ideal_timestamps, '--', color='black', linewidth=.5, label=ideal_line_label)- ax_frame_duration.legend()-- plt.show()- plt.pause(0.1)- f=9 #put a breakpoint here to view/interact with figure if `plt.ion()` is active+#%% connect to all available cameras++camera_found_bool = True+camera_list = []+this_port_number = -1+while camera_found_bool:+ this_port_number+=1+ camera = OpenCVCamera(port_number=this_port_number)+ camera_found_bool = camera.connect()+ if camera_found_bool:+ camera_list.append(camera)++number_of_cameras = len(camera_list)++global thread_exit_event+thread_exit_event = threading.Event()+loop_count = 0+camera_thread_list = [None]*len(camera_list)+incoming_frames_queue = queue.Queue()++#%% set some things up before we start streaming+# setting up to grab incoming timestamps etc+dict_of_timestamp_lists = {}+dict_of_fps_mean_lists = {}+dict_of_fps_std_lists = {}+for this_camera in camera_list:+ dict_of_timestamp_lists[this_camera.name] = []+ dict_of_fps_mean_lists[this_camera.name] = []+ dict_of_fps_std_lists[this_camera.name] = []++# #create cv2 named windows where we'll display incoming frames+cam_window_name_dict = {}+display_image_scale_factor = 2+scaled_image_width = int(camera_list[0].resolution_width/display_image_scale_factor)+scaled_image_height = int(camera_list[0].resolution_height/display_image_scale_factor)+window_buffer = 50+for this_camera in camera_list:+ #tile windows across monitor+ window_location_x = this_camera.port_number*scaled_image_width+ window_location_y=0+ if window_location_x+scaled_image_width > 1920:+ window_location_x -= 1920+ window_location_y = scaled_image_height+window_buffer+ elif window_location_x+scaled_image_width > 1920*2:+ window_location_x -= 1920*2+ window_location_y = scaled_image_height*2+ print('Creating window for {} at ({},{})'.format(this_camera.name, str(window_location_x), str(window_location_y)))+ cam_window_name_dict[this_camera.name] = create_video_viewing_window(this_camera.name, (window_location_x, window_location_y))++# multi_cam_widow_name = create_video_viewing_window('MultiCamera')++# #were going to scale down each incoming camera image and tile it, so we only have to display one big thing (which is faster than displaying many small things)+# multi_image_number_of_tiles_per_side = int(np.ceil(np.sqrt(number_of_cameras)))+# num_extra_spots_in_multi_image = multi_image_number_of_tiles_per_side - number_of_cameras+# display_image_scale_factor = 2+# scaled_image_width = int(camera_list[0].resolution_width/display_image_scale_factor)+# scaled_image_height = int(camera_list[0].resolution_height/display_image_scale_factor)++# scaled_image_black_frame = np.zeros((scaled_image_height, scaled_image_width, 3))+# multi_image_holder_list_of_lists = []++# for iter in range(multi_image_number_of_tiles_per_side):+# blank_row = [scaled_image_black_frame.copy() for each in range(multi_image_number_of_tiles_per_side)]+# multi_image_holder_list_of_lists.append(blank_row)++# each_cameras_multi_image_address = []+# for this_cam_num in range(number_of_cameras):+# this_cam_row = int(np.floor(this_cam_num/multi_image_number_of_tiles_per_side))+# this_cam_col = this_cam_num - this_cam_row*multi_image_number_of_tiles_per_side+# each_cameras_multi_image_address.append((this_cam_row, this_cam_col))++++#%%+###########################################################################################+###########################################################################################+##+## Start the camera threads (THIS IS WHERE THE MAGIC HAPPENS)+##+###########################################################################################+###########################################################################################++for this_camera in camera_list:+ camera_thread_list[this_camera.port_number] = start_camera_thread(this_camera, incoming_frames_queue)++#+# start the MainThread loop+approx_start_time = time.time()+end_after_seconds = 20 #or on pressing ESC+qsize = []+while not thread_exit_event.is_set():+ time_elapsed = time.time() - approx_start_time+ this_qsize = incoming_frames_queue.qsize()+ qsize.append(this_qsize)+ print('Time remaining: {:3.2f} - Queueue size: {:3.0f} '.format(end_after_seconds - time_elapsed, this_qsize))+ if time_elapsed > end_after_seconds:+ thread_exit_event.set()+ cv2.destroyAllWindows()++ if not incoming_frames_queue.empty():+ this_frame_tuple = incoming_frames_queue.get() #<- grab the most recent frame_tuple and do stuff with it+ this_success_bool, this_camera_number, this_camera_name, this_image, this_timestamp_ns = this_frame_tuple+ this_timestamp_sec = this_timestamp_ns/1e9+ dict_of_timestamp_lists[this_camera_name].append(this_timestamp_sec)++ elif qsize[-1] < 10: #only display images if queue isn't over filled+ # this_cam_address = each_cameras_multi_image_address[this_camera_number]+ # print(this_cam_address)+ # resized_image = cv2.resize(this_image, (scaled_image_width, scaled_image_height))+ # multi_image_holder_list_of_lists[this_cam_address[0]][this_cam_address[1]] = resized_image++ # each_row_of_multi_im = []+ # for this_row in range(multi_image_number_of_tiles_per_side):+ # each_row_of_multi_im.append(np.hstack(multi_image_holder_list_of_lists[this_row]))+ # this_multi_im = np.vstack(each_row_of_multi_im)+ # this_multi_im = cv2.normalize(src=this_multi_im, dst=None, alpha=0, beta=255, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_8U)+ # display_image_on_its_cv2_named_window(multi_cam_widow_name, this_multi_im)+ image_height, image_width, image_channels = this_image.shape+ scale_height = int(image_height/display_image_scale_factor)+ scale_width = int(image_width/display_image_scale_factor)+ display_image_on_its_cv2_named_window(cam_window_name_dict[this_camera_name], cv2.resize(this_image, (scale_width, scale_height)))++#%%++list_lengths = [len(this_thing) for this_thing in dict_of_timestamp_lists ]+for this_camera in camera_list:+ this_cam_timestamps = np.array(dict_of_timestamp_lists[this_camera.name])++number_of_frames = np.min([len(dict_of_timestamp_lists[this_key]) for this_key in dict_of_timestamp_lists ]) #truncate all timestamp lists to match frame count of shortest+out_timestamp_dict = {}++for this_cam_num, this_camera in enumerate(camera_list):+ out_timestamp_dict[this_camera.name] = np.array(dict_of_timestamp_lists[this_camera.name][:number_of_frames])+out_dataframe = pd.DataFrame(out_timestamp_dict)+out_dataframe.to_csv('~/timestamps.csv')++#%% Plot up them bad bois+# where 'bad boi' is defined as a timestamp from each camera+# plt.ion() #<- allows for fiddling, but need to set breakpoint and call plt.show() to see the plots+figure = plt.figure(3214, figsize=(18,5))+ax_timestamps = figure.add_subplot(141)+ax_timestamps.set_xlabel('frame number')+ax_timestamps.set_ylabel('timestamp (unix epoch, sec)')++ax_frame_duration = figure.add_subplot(142)+ax_frame_duration.set_xlabel('frame number')+ax_frame_duration.set_ylabel('framerate (1/np.diff(timestamp, frames/sec)')+ax_frame_duration.set_ylim(0, .1)++ax_histogram = figure.add_subplot(143)+ax_histogram.set_xlabel('bins (fps)')+ax_histogram.set_ylabel('histogram count(out of {} frames'.format(number_of_frames))++ax_qsize = figure.add_subplot(144)+ax_qsize.set_xlabel('frame/loop count')+ax_qsize.set_ylabel('incoming frames queueue size')+ax_qsize.plot(qsize, '-',linewidth=.5, alpha=.5)+ax_qsize.plot(qsize, 'k.')++for this_camera in camera_list:+ this_cam_timestamps = out_timestamp_dict[this_camera.name]+ this_cam_frame_duration = np.diff(out_timestamp_dict[this_camera.name])+ this_cam_framerate = 1/this_cam_frame_duration++ ax_timestamps.plot(this_cam_timestamps, 'o',markersize=1, alpha=.5, label=this_camera.name)+ ax_frame_duration.plot(this_cam_frame_duration, '-o', markersize=1, alpha=.5, label=this_camera.name)+ ax_histogram.hist(this_cam_framerate, bins=np.arange(0,50,1), alpha=.5,label=this_camera.name)++#reference lines (i.e. 'ideal' numbers)+ideal_framerate = 30+approx_start_time = this_cam_timestamps[0] #just for the last camera, but it's fine for this visualization+ideal_end_time = approx_start_time + number_of_frames/ideal_framerate+ideal_timestamps = np.linspace(approx_start_time, ideal_end_time, number_of_frames)++ideal_line_label = 'ideal for {}fps'.format(ideal_framerate)+ax_frame_duration.plot([ 0, number_of_frames], [1/ideal_framerate, 1/ideal_framerate],'--', label=ideal_line_label, color='black')+ax_histogram.vlines(ideal_framerate, 0, number_of_frames, linestyles='dashed', label=ideal_line_label)+ax_timestamps.plot(ideal_timestamps, '--', color='black', label=ideal_line_label)+ax_frame_duration.legend()++plt.show()+plt.pause(0.1)#put a breakpoint here to view/interact with figure if `plt.ion()` is active+