在输出视频流上计算FPS的麻烦

Question

我已经使用TensorFlow和打开CV实现了实时对象检测，如本文和此GitHub repo中所述。

对象检测效果很好，但是我在实现FPS计算器方面遇到困难。

这是我写的代码：

import datetime

class FramesPerSecond:
    def __init__(self):
    self._startTime = None
    self._currentTime = None
    self._total_number_of_frames = 0
def start(self):
    # start timer
    self._startTime = datetime.datetime.now()
    return self
def stop(self):
    # stop times
    self._currentTime = datetime.datetime.now()
def update_frames(self):
    self._total_number_of_frames = self._total_number_of_frames + 1
def elapsed_time(self):
    return (datetime.datetime.now() - self._startTime).total_seconds()
def current_fps(self):
    return self._total_number_of_frames / self.elapsed_time()

而不是在视频流的持续时间内保持相同的值，而是返回在整个流中增加的值。

Answer 1

current_fps函数返回一个随时间增加的值。基本上，分子将永远增加，而分母将保持在范围内，从而导致值随着时间的推移增加。

为了避免此问题，我们必须允许经过的时间函数随着帧的增加而扩展，或者使用其他常数来测量我们的FPS。让我们继续使用后者，保持单独的计数器变量。

因此，您的代码可以像这样重写：

import time  # The time module is easier for our purpose.

class FramesPerSecond:
    def __init__(self):
        self.startTime = None
        self.total_number_of_frames = 0
        self.counter = 0
        self.frameRate = 1  # The number of seconds to wait for each measurement.
    def start(self):
        self.startTime = time.time()  # Returns a UNIX timestamp.
    def update_frames(self):
        self.total_number_of_frames += 1
    def elapsed_time(self):
        return time.time() - self.startTime
    def current_fps(self):
        return self.counter / self.elapsed_time()  # We are measuring against the counter instead of no. of frames.

fps = FramesPerSecond()
fps.start()
while True:  # Your main loop.
    # Rest of your code here #
    fps.update_frames()
    fps.counter += 1  # Count will increase until the if condition executes.
    if fps.elapsed_time() > fps.frameRate:  # We measure the fps only after 1 second has passed.
        print("FPS: ", fps.current_fps()) # The frames per second. 
        fps.counter = 0  # reset the counter for next iteration.
        fps.start()  # reset the start time.

希望这会有所帮助！

Answer 2

您可以尝试这样的东西...尽管我不确定您是否完全想要此或其他东西...

import datetime
class FPS:
    def __init__(self):
        ...
        ...
        self._elapsed = 0
        self._fps = 0
    def start(self, ttl):
        self._startTime = datetime.datetime.now()
        self._currentTime = datetime.datetime.now() + datetime.timedelta(seconds=ttl)
        # print('counter started at --> ', self._startTime)
        # print('counter stopped at --> ', self._currentTime)
        return self._startTime, self._currentTime
    def elapsed_time(self):
        self._elapsed = (self._currentTime - self._startTime).total_seconds()
        return self._elapsed
    def update_frames(self):
        curr, start = self._currentTime, self._startTime
        if curr != start and curr != 0:
            print('updatin frames ...')
            for i in range(int(self._elapsed) + 1):
                self._total_number_of_frames += i
        else:
            self._total_number_of_frames += 1
        self.current_fps()
        print("frames: {}, fps: {}, time spent: {}".format(
            self._total_number_of_frames,
            self._fps,
            self._elapsed
        ))
    def current_fps(self):
        self._fps = self._total_number_of_frames / self._elapsed
        return self._fps

if __name__ == '__main__':
    fps = FPS()
    ttl = int(input('time-to-live in seconds: '))
    fps.start(ttl)
    fps.elapsed_time()
    fps.update_frames()