blakeblackshear.frigate/frigate/process_clip.py

290 lines
9.1 KiB
Python
Raw Normal View History

2020-09-07 19:17:42 +02:00
import datetime
2020-12-12 16:12:15 +01:00
import json
import logging
2020-09-07 19:17:42 +02:00
import multiprocessing as mp
2020-12-12 16:12:15 +01:00
import os
import subprocess as sp
import sys
from unittest import TestCase, main
import click
2020-09-07 19:17:42 +02:00
import cv2
2020-12-12 16:12:15 +01:00
import numpy as np
from frigate.config import FRIGATE_CONFIG_SCHEMA, FrigateConfig
from frigate.edgetpu import LocalObjectDetector
from frigate.motion import MotionDetector
2020-09-07 19:17:42 +02:00
from frigate.object_processing import COLOR_MAP, CameraState
2020-12-12 16:12:15 +01:00
from frigate.objects import ObjectTracker
2021-02-17 14:23:32 +01:00
from frigate.util import (
DictFrameManager,
EventsPerSecond,
SharedMemoryFrameManager,
draw_box_with_label,
)
from frigate.video import capture_frames, process_frames, start_or_restart_ffmpeg
2020-12-12 16:12:15 +01:00
logging.basicConfig()
logging.root.setLevel(logging.DEBUG)
logger = logging.getLogger(__name__)
2021-02-17 14:23:32 +01:00
2020-12-12 16:12:15 +01:00
def get_frame_shape(source):
2021-02-17 14:23:32 +01:00
ffprobe_cmd = " ".join(
[
"ffprobe",
"-v",
"panic",
"-show_error",
"-show_streams",
"-of",
"json",
'"' + source + '"',
]
)
2020-12-12 16:12:15 +01:00
p = sp.Popen(ffprobe_cmd, stdout=sp.PIPE, shell=True)
(output, err) = p.communicate()
p_status = p.wait()
info = json.loads(output)
2021-02-17 14:23:32 +01:00
video_info = [s for s in info["streams"] if s["codec_type"] == "video"][0]
if video_info["height"] != 0 and video_info["width"] != 0:
return (video_info["height"], video_info["width"], 3)
2020-12-12 16:12:15 +01:00
# fallback to using opencv if ffprobe didnt succeed
video = cv2.VideoCapture(source)
ret, frame = video.read()
frame_shape = frame.shape
video.release()
return frame_shape
2020-09-07 19:17:42 +02:00
2021-02-17 14:23:32 +01:00
class ProcessClip:
2020-12-12 16:12:15 +01:00
def __init__(self, clip_path, frame_shape, config: FrigateConfig):
2020-09-07 19:17:42 +02:00
self.clip_path = clip_path
2021-02-17 14:23:32 +01:00
self.camera_name = "camera"
2020-12-12 16:12:15 +01:00
self.config = config
2021-02-17 14:23:32 +01:00
self.camera_config = self.config.cameras["camera"]
2020-12-12 16:12:15 +01:00
self.frame_shape = self.camera_config.frame_shape
2021-02-17 14:23:32 +01:00
self.ffmpeg_cmd = [
c["cmd"] for c in self.camera_config.ffmpeg_cmds if "detect" in c["roles"]
][0]
2020-12-12 16:12:15 +01:00
self.frame_manager = SharedMemoryFrameManager()
2020-09-07 19:17:42 +02:00
self.frame_queue = mp.Queue()
self.detected_objects_queue = mp.Queue()
self.camera_state = CameraState(self.camera_name, config, self.frame_manager)
def load_frames(self):
fps = EventsPerSecond()
skipped_fps = EventsPerSecond()
2021-02-17 14:23:32 +01:00
current_frame = mp.Value("d", 0.0)
frame_size = (
self.camera_config.frame_shape_yuv[0]
* self.camera_config.frame_shape_yuv[1]
)
ffmpeg_process = start_or_restart_ffmpeg(
self.ffmpeg_cmd, logger, sp.DEVNULL, frame_size
)
capture_frames(
ffmpeg_process,
self.camera_name,
self.camera_config.frame_shape_yuv,
self.frame_manager,
self.frame_queue,
fps,
skipped_fps,
current_frame,
)
2020-09-07 19:17:42 +02:00
ffmpeg_process.wait()
ffmpeg_process.communicate()
2021-02-17 14:23:32 +01:00
def process_frames(self, objects_to_track=["person"], object_filters={}):
2020-09-07 19:17:42 +02:00
mask = np.zeros((self.frame_shape[0], self.frame_shape[1], 1), np.uint8)
mask[:] = 255
2021-02-17 14:23:32 +01:00
motion_detector = MotionDetector(
self.frame_shape, mask, self.camera_config.motion
)
2020-09-07 19:17:42 +02:00
2021-02-17 14:23:32 +01:00
object_detector = LocalObjectDetector(labels="/labelmap.txt")
object_tracker = ObjectTracker(self.camera_config.detect)
2020-12-12 16:12:15 +01:00
process_info = {
2021-02-17 14:23:32 +01:00
"process_fps": mp.Value("d", 0.0),
"detection_fps": mp.Value("d", 0.0),
"detection_frame": mp.Value("d", 0.0),
2020-12-12 16:12:15 +01:00
}
2020-09-07 19:17:42 +02:00
stop_event = mp.Event()
2020-12-12 16:12:15 +01:00
model_shape = (self.config.model.height, self.config.model.width)
2020-09-07 19:17:42 +02:00
2021-02-17 14:23:32 +01:00
process_frames(
self.camera_name,
self.frame_queue,
self.frame_shape,
model_shape,
self.frame_manager,
motion_detector,
object_detector,
object_tracker,
self.detected_objects_queue,
process_info,
objects_to_track,
object_filters,
mask,
stop_event,
exit_on_empty=True,
)
def top_object(self, debug_path=None):
2020-09-07 19:17:42 +02:00
obj_detected = False
top_computed_score = 0.0
2021-02-17 14:23:32 +01:00
2020-12-12 16:12:15 +01:00
def handle_event(name, obj, frame_time):
2020-09-07 19:17:42 +02:00
nonlocal obj_detected
nonlocal top_computed_score
2020-12-12 16:12:15 +01:00
if obj.computed_score > top_computed_score:
top_computed_score = obj.computed_score
if not obj.false_positive:
2020-09-07 19:17:42 +02:00
obj_detected = True
2021-02-17 14:23:32 +01:00
self.camera_state.on("new", handle_event)
self.camera_state.on("update", handle_event)
while not self.detected_objects_queue.empty():
(
camera_name,
frame_time,
current_tracked_objects,
motion_boxes,
regions,
) = self.detected_objects_queue.get()
2020-09-07 19:17:42 +02:00
if not debug_path is None:
2021-02-17 14:23:32 +01:00
self.save_debug_frame(
debug_path, frame_time, current_tracked_objects.values()
)
self.camera_state.update(
frame_time, current_tracked_objects, motion_boxes, regions
)
2020-09-07 19:17:42 +02:00
self.frame_manager.delete(self.camera_state.previous_frame_id)
2021-02-17 14:23:32 +01:00
return {"object_detected": obj_detected, "top_score": top_computed_score}
2020-09-07 19:17:42 +02:00
def save_debug_frame(self, debug_path, frame_time, tracked_objects):
2021-02-17 14:23:32 +01:00
current_frame = cv2.cvtColor(
self.frame_manager.get(
f"{self.camera_name}{frame_time}", self.camera_config.frame_shape_yuv
),
cv2.COLOR_YUV2BGR_I420,
)
2020-09-07 19:17:42 +02:00
# draw the bounding boxes on the frame
for obj in tracked_objects:
thickness = 2
2021-02-17 14:23:32 +01:00
color = (0, 0, 175)
2020-09-07 19:17:42 +02:00
2021-02-17 14:23:32 +01:00
if obj["frame_time"] != frame_time:
2020-09-07 19:17:42 +02:00
thickness = 1
2021-02-17 14:23:32 +01:00
color = (255, 0, 0)
2020-09-07 19:17:42 +02:00
else:
2021-02-17 14:23:32 +01:00
color = (255, 255, 0)
2020-09-07 19:17:42 +02:00
# draw the bounding boxes on the frame
2021-02-17 14:23:32 +01:00
box = obj["box"]
draw_box_with_label(
current_frame,
box[0],
box[1],
box[2],
box[3],
obj["id"],
f"{int(obj['score']*100)}% {int(obj['area'])}",
thickness=thickness,
color=color,
)
2020-09-07 19:17:42 +02:00
# draw the regions on the frame
2021-02-17 14:23:32 +01:00
region = obj["region"]
draw_box_with_label(
current_frame,
region[0],
region[1],
region[2],
region[3],
"region",
"",
thickness=1,
color=(0, 255, 0),
)
cv2.imwrite(
f"{os.path.join(debug_path, os.path.basename(self.clip_path))}.{int(frame_time*1000000)}.jpg",
current_frame,
)
2020-09-07 19:17:42 +02:00
@click.command()
@click.option("-p", "--path", required=True, help="Path to clip or directory to test.")
2021-02-17 14:23:32 +01:00
@click.option("-l", "--label", default="person", help="Label name to detect.")
2020-09-07 19:17:42 +02:00
@click.option("-t", "--threshold", default=0.85, help="Threshold value for objects.")
@click.option("-s", "--scores", default=None, help="File to save csv of top scores")
2020-09-07 19:17:42 +02:00
@click.option("--debug-path", default=None, help="Path to output frames for debugging.")
def process(path, label, threshold, scores, debug_path):
2020-09-07 19:17:42 +02:00
clips = []
if os.path.isdir(path):
files = os.listdir(path)
files.sort()
clips = [os.path.join(path, file) for file in files]
2021-02-17 14:23:32 +01:00
elif os.path.isfile(path):
2020-09-07 19:17:42 +02:00
clips.append(path)
2020-12-12 16:12:15 +01:00
json_config = {
2021-02-17 14:23:32 +01:00
"mqtt": {"host": "mqtt"},
"cameras": {
"camera": {
"ffmpeg": {
"inputs": [
{
"path": "path.mp4",
"global_args": "",
"input_args": "",
"roles": ["detect"],
}
2020-12-12 16:12:15 +01:00
]
},
2021-02-17 14:23:32 +01:00
"height": 1920,
"width": 1080,
2020-09-07 19:17:42 +02:00
}
2021-02-17 14:23:32 +01:00
},
2020-09-07 19:17:42 +02:00
}
results = []
for c in clips:
2020-12-12 16:12:15 +01:00
logger.info(c)
2020-09-07 19:17:42 +02:00
frame_shape = get_frame_shape(c)
2021-02-17 14:23:32 +01:00
json_config["cameras"]["camera"]["height"] = frame_shape[0]
json_config["cameras"]["camera"]["width"] = frame_shape[1]
json_config["cameras"]["camera"]["ffmpeg"]["inputs"][0]["path"] = c
2020-12-12 16:12:15 +01:00
config = FrigateConfig(config=FRIGATE_CONFIG_SCHEMA(json_config))
2020-09-07 19:17:42 +02:00
process_clip = ProcessClip(c, frame_shape, config)
process_clip.load_frames()
2020-12-12 16:12:15 +01:00
process_clip.process_frames(objects_to_track=[label])
2020-09-07 19:17:42 +02:00
results.append((c, process_clip.top_object(debug_path)))
2020-09-07 19:17:42 +02:00
if not scores is None:
2021-02-17 14:23:32 +01:00
with open(scores, "w") as writer:
for result in results:
writer.write(f"{result[0]},{result[1]['top_score']}\n")
2020-09-07 19:17:42 +02:00
2021-02-17 14:23:32 +01:00
positive_count = sum(1 for result in results if result[1]["object_detected"])
print(
f"Objects were detected in {positive_count}/{len(results)}({positive_count/len(results)*100:.2f}%) clip(s)."
)
if __name__ == "__main__":
2020-12-12 16:12:15 +01:00
process()