A blog by Devendra Tewari
Picamera2 is a Python library based on libcamera that replaces Picamera.
This post discusses sample code that captures full-resolution (5MP) still images from a OV5604 (camera module V1) sensor and its performance, on a Raspberry Pi 4 with both 32-bit and 64-bit Raspberry Pi OS. Raspberry Pi OS already has all the dependencies required to run the sample code. /etc/os-release shows the Raspberry Pi OS version to be 11 (bullseye). apt info libcamera0 shows its version to be 0~git20230124+9b860a66-1.
Power supplies can significantly influence performance—always use the original 3A power adapter, or better. SD cards can also influence performance, use fio to benchmark and compare different cards
sudo apt install fio
fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=32M --readwrite=randrw --rwmixread=75
The code below sequentially captures twenty full-resolution JPEG images using capture_file function
from picamera2 import Picamera2, Preview
import time
picam2 = Picamera2()
camera_config = picam2.create_still_configuration()
picam2.configure(camera_config)
picam2.start()
time.sleep(2)
counter = 0
start = time.time()
while 1:
print(f"capture {counter}")
picam2.capture_file(f"test{counter}.jpg")
counter += 1
if counter == 20:
break
end = time.time()
print(end - start)
It takes 5.64 seconds to run on 32-bit Raspberry Pi OS and 3.85 seconds on 64-bit.
The code below is quite similar to the one above, but it makes a Python Image Library (PIL) image from the capture buffer using make_image helper function, then calls save
from picamera2 import Picamera2, Preview
import time
picam2 = Picamera2()
camera_config = picam2.create_still_configuration()
picam2.configure(camera_config)
picam2.start()
time.sleep(2)
counter = 0
start = time.time()
while 1:
print(f"capture {counter}")
(buffer, ), metadata = picam2.capture_buffers(["main"])
img = picam2.helpers.make_image(buffer, camera_config["main"])
picam2.helpers.save(img, metadata, f"test{counter}.jpg")
counter += 1
if counter == 20:
break
end = time.time()
print(end - start)
It takes 6.36 seconds to run on 32-bit Raspberry Pi OS and 3.82 seconds on 64-bit.
This code changes the previous code to use a separate thread to call make_image helper function, then call save
from picamera2 import Picamera2, Preview
import time
import threading
picam2 = Picamera2()
camera_config = picam2.create_still_configuration(buffer_count=2)
picam2.configure(camera_config)
picam2.start()
time.sleep(2)
counter = 0
def write_jpg(buffer):
img = picam2.helpers.make_image(buffer, camera_config["main"])
picam2.helpers.save(img, metadata, f"test{counter}.jpg")
start = time.time()
while 1:
print(f"capture {counter}")
(buffer, ), metadata = picam2.capture_buffers(["main"])
t = threading.Thread(target=write_jpg, args=(buffer,))
t.start()
counter += 1
if counter == 20:
break
for t in threading.enumerate():
if t != threading.current_thread() and t.daemon is False:
t.join()
end = time.time()
print(end - start)
It takes 2.18 seconds to run on 32-bit Raspberry Pi OS and 1.37 seconds on 64-bit. Comparing this with result from previous sample demonstrates that optimizing make_image is critical to improving performance.
This code is similar to the one before but instead of writing to file, it writes data out to a TCP socket
from picamera2 import Picamera2, Preview
import time
import threading
import socket
picam2 = Picamera2()
camera_config = picam2.create_still_configuration(buffer_count=2)
picam2.configure(camera_config)
picam2.start()
time.sleep(2)
counter = 0
def write_jpg(buffer):
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.connect(("192.168.68.60", 8001))
out = sock.makefile("wb")
img = picam2.helpers.make_image(buffer, camera_config["main"])
picam2.helpers.save(img, metadata, out, format='jpeg')
start = time.time()
while 1:
print(f"capture {counter}")
(buffer, ), metadata = picam2.capture_buffers(["main"])
t = threading.Thread(target=write_jpg, args=(buffer,))
t.start()
counter += 1
if counter == 20:
break
for t in threading.enumerate():
if t != threading.current_thread() and t.daemon is False:
t.join()
end = time.time()
print(end - start)
It takes 2.18 seconds to run on 32-bit Raspberry Pi OS and 1.37 seconds on 64-bit, same as the previous sample.
To receive at the other end
import socket
import threading
from datetime import datetime
def save_jpeg(args):
conn = args[0]
with open(f"{datetime.utcnow().strftime('%Y%m%d-%H%M%S_%f')[:-3]}.jpg", "wb") as f:
while 1:
data = conn.recv(1024)
if data:
f.write(data)
else:
break
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
sock.bind(("0.0.0.0", 8001))
sock.listen()
while 1:
conn = sock.accept()
t = threading.Thread(target=save_jpeg, args=(conn,))
t.start()
This sample code recovers a raw image buffer and calls save_dng helper to save a raw image
from picamera2 import Picamera2, Preview
import time
picam2 = Picamera2()
camera_config = picam2.create_still_configuration(raw={})
picam2.configure(camera_config)
picam2.start()
time.sleep(2)
counter = 0
start = time.time()
while 1:
print(f"capture {counter}")
(buffer, ), metadata = picam2.capture_buffers(["raw"])
picam2.helpers.save_dng(buffer, metadata, camera_config["raw"], f"test{counter}.dng")
counter += 1
if counter == 20:
break
end = time.time()
print(end - start)
It takes 17.82 seconds to run on 32-bit Raspberry Pi OS and 10.46 seconds on 64-bit.