I know the title is a bit vague but I'm not sure how else to describe it.
CentOS with ffmpeg + OpenCV 2.4.9. I'm working on a simple motion detection system which uses a stream from an IP camera (h264).
Once in a while the stream hiccups and throws in a "bad frame" (see pic-bad.png link below). The problem is, these frames vary largely from the previous frames and causes a "motion" event to get triggered even though no actual motion occured.
The pictures below will explain the problem.
Good frame (motion captured):
Bad frame (no motion, just a broken frame):
The bad frame gets caught randomly. I guess I can make a bad frame detector by analyzing (looping) through the pixels going down from a certain position to see if they are all the same, but I'm wondering if there is any other, more efficient, "by the book" approach to detecting these types of bad frames and just skipping over them.
Thank You!
EDIT UPDATE:
The frame is grabbed using a C++ motion detection program via cvQueryFrame(camera); so I do not directly interface with ffmpeg, OpenCV does it on the backend. I'm using the latest version of ffmpeg compiled from git source. All of the libraries are also up to date (h264, etc, all downloaded and compiled yesterday). The data is coming from an RTSP stream (ffserver). I've tested over multiple cameras (dahua 1 - 3 MP models) and the frame glitch is pretty persistent across all of them, although it doesn't happen continuously, just once on a while (ex: once every 10 minutes).
What comes to my mind in first approach is to check dissimilarity between example of valid frame and the one we are checking by counting the pixels that are not the same. Dividing this number by the area we get percentage which measures dissimilarity. I would guess above 0.5 we can say that tested frame is invalid because it differs too much from the example of valid one.
This assumption is only appropriate if you have a static camera (it does not move) and the objects which can move in front of it are not in the shortest distance (depends from focal length, but if you have e.g. wide lenses so objects should not appear less than 30 cm in front of camera to prevent situation that objects "jumps" into a frame from nowhere and has it size bigger that 50% of frame area).
Here you have opencv function which does what I said. In fact you can adjust dissimilarity coefficient more large if you think motion changes will be more rapid. Please notice that first parameter should be an example of valid frame.
bool IsBadFrame(const cv::Mat &goodFrame, const cv::Mat &nextFrame) {
// assert(goodFrame.size() == nextFrame.size())
cv::Mat g, g2;
cv::cvtColor(goodFrame, g, CV_BGR2GRAY);
cv::cvtColor(nextFrame, g2, CV_BGR2GRAY);
cv::Mat diff = g2 != g;
float similarity = (float)cv::countNonZero(diff) / (goodFrame.size().height * goodFrame.size().width);
return similarity > 0.5f;
}
You do not mention if you use ffmpeg command line or libraries, but in the latter case you can check the bad frame flag (I forgot its exact description) and simply ignore those frames.
remove waitKey(50) or change it to waitKey(1). I think opencv does not spawn a new thread to perform capture. so when there is a pause, it confuses the buffer management routines, causing bad frames..maybe?
I have dahua cameras and observed that with higher delay, bad frames are observed. And they go away completely with waitKey(1). The pause does not necessarily need to come from waitKey. Calling routines also cause such pauses and result in bad frames if they are taking long enough.
This means that there should be minimum pause between consecutive frame grabs.the solution would be to use two threads to perform capture and processing separately.
Related
I have a custom USB camera with a custom driver on a custom board Nvidia Jetson TX2 that is not detected through openpose examples. I access the data using GStreamer custom source. I currently pull frames into a CV mat, color convert them and feed into OpenPose on a per picture basis, it works fine but 30 - 40% slower than a comparable video stream from a plug and play camera. I would like to explore things like tracking that is available for streams since Im trying to maximize the fps. I believe the stream feed is superior due to better (continuous) use of the GPU.
In particular the speedup would come at confidence expense and would be addressed later. 1 frame goes through pose estimation and 3 - 4 subsequent frames are just tracking the object with decreasing confidence levels. I tried that on a plug and play camera and openpose example and the results were somewhat satisfactory.
The point where I stumbled is that I can put the video stream into CV VideoCapture but I do not know, however, how to provide the CV video capture to OpenPose for processing.
If there is a better way to do it, I am happy to try different things but the bottom line is that the custom camera stays (I know ;/). Solutions to the issue described or different ideas are welcome.
Things I already tried:
Lower resolution of the camera (the camera crops below certain res instead of binning so cant really go below 1920x1080, its a 40+ MegaPixel video camera by the way)
use CUDA to shrink the image before feeding it to OpenPose (the shrink + pose estimation time was virtually equivalent to the pose estimation on the original image)
since the camera view is static, check for changes between frames, crop the image down to the area that changed and run pose estimation on that section (10% speedup, high risk of missing something)
I am trying to capture images from several cameras using the cameras driver,OpenCV and C++. My goal is to get as many FPS as possible, and to this end I have found saving the images in the hard drive to be the slowest operation. In order to speed up the process, I am doing each saving in separate threads. Problem is, I still have to wait for the saving to be complete to avoid the captured image being overwritten. Doing this provides good results, but for unknown reasons every 30-40 frames the speed is 10x higher.
I am addressing this by creating a ring buffer where I store the images, as these sudden drops in write speed are very short. I have obtained very good results using this approach, but unfortunately for more than 3 cameras the camera driver can't handle the stress and my program halts, waiting for the first image of the 4th camera to be saved. I checked and it's not the CPU, as 3 cameras + a thread writing random data in the disk works fine.
Now, seeing how using opencv reduced the stress on the camera driver, I would like to create a OpenCV mat buffer to hold the images while they are saved without my camera overwritting them (well, not until the buffer has done a whole lap, which I will make sure won't happen).
I know I can do
cv::Mat colorFrame(cv::Size(width, height),CV_8UC3,pointerToMemoryOfCamera);
to initialize a frame from the memory written by the camera. This does not solve my problem, as it will only point to the data, and the moment the camera overwrites it, it will corrupt the image saved.
How do I create a matrix with a given size and type, and then copy the contents of the memory to this matrix?
You need to create a deep copy. You can use clone:
cv::Mat colorFrame = cv::Mat(height, width, CV_8UC3, pointerToMemoryOfCamera).clone();
You can also speed up the process of saving the images using matwrite and matread functions.
I have a program that reads from video camera, and finds the maximum color intensity of all the frames. I noticed that there is a dashed line effect when I streak a laser pointer across the camera viewing area.
I think this is because the fps of the camera is less than the shutter speed of the camera. or that the waitkey function prevents the camera to read during that period.
Is there a way to remove this effect so that if I streak a laser pointer across the camera view it leaves a continuous line?
EDIT: I also need the results in real time if that is possible.
here is the simplified code below:
while 1:
ret, view = vid.read()
outputImg = processImage(view)
cv2.waitkey(1)
imshow(outputImg)
You should try to first save the video and then process it later.
while 1:
ret, view = vid.read()
outputImg = processImage(view)
cv2.waitkey(1)
imshow(outputImg)
Your code captures a frame and then processes it using your processImage function and then waits in cv2.waitkey(1) and then displays the processed frame, and finally, again reads the next frame. So there is a time lapse between reading of the two frames. You can save it like this:
while 1:
ret, frame = vid.read()
out.write(frame)
While shiva's suggestion will probably improve your results, it will still be unpredictable, and the behaviour will greatly depend on the capabilities of the camera and your ability to control it. I would expect the gaps to be there.
Instead, I would make an attempt at correcting this:
Find the endpoints of each streak in a frame.
Find the two adjacent endpoints in adjacent frames.
Draw artificial laser line connecting those two points. (Perhaps by using a sample taken from both ends and interpolating between them)
If for some reason you can't interpolate, but still need to be able to do the processing in near-realtime, consider putting the image acquisition into a separate thread. Set this thread to high priority, to minimise the gaps between the frames.
This thread will repeatedly:
Acquire frame from camera.
Insert the frame into a synchronized queue.
Then, create a processing thread (this could be your main thread as well), which will take the frames from the synchronized queue, and do whatever processing is necessary.
To achieve the best results, you should use a camera that is able to run in "freerun" mode, where the camera automatically triggers itself, and sends a stream of data to the computer to process.
This might not be possible to do directly with OpenCV and a regular webcam. Some lower-end industrial camera, either a GigEVision or USB3 model might be more appropriate. (AVT Manta, Basler Scout, etc.) You may be able to find some good deals on EBay if the cost is out of your price range.
You would most likely need to use a special API to control the camera and acquire the frames, but the rest of the processing would be done using OpenCV.
First of all, I understand this question has been asked several times at here.
FindChessboardCorners cannot detect chessboard on very large images by long focal length lens
Opencv corners detection for High resolution images
However, my situation is a little bit different.
My first experiment is retrieving sequential images of 3264 x 2448 from a Webcam which supports resolution this high, and uses findChessboardCorners to detect the corners on a pattern I placed.
Gladly it works! So I move to next experiment.(See success cases below, I cropped them)
This time I try to project a pattern from a projector of my own to a clean board and detect it, Sadly I failed at here.(Example below, 2592 x 1944)
The two experiments retrieves similar images(I think so), but how come one succeeds and the other one won't? Especially the succeed one has the highest resolution.
I also tried adjust the size of the pattern that projector projects, didn't work.
Adjust the distance of the board, didn't work.
Adjust the camera settings, from lighter to darker, didn't work.
By the way, I suppose the resolution I choose affects the camera intrinsic parameters, so "resize" the image shouldn't be a good idea right? Since I require parameters under high resolution.
I am attempting to use a straightforward motion detection code to detect movement from a camera. I'm using the OpenCV library and I have some code that takes the difference between two frames to detect a change.
I have the difference frame working just fine and it's black when no motion is present.
The problem is how now i can detect that blackness to stop recording or no darkness to begin recording frames.
Thank u all.
A very simple thing to do is to sum the entire diff image into an integer. If that sum is above a threshold you have movement. Then you can use a second threshold and when the sum is below that limit you stopped having movement.
You can also make the threshold only change the program state if some elapsed time has occurred since the last threshold. i.e. after movement is detected you don't check for lack of movement for 10 seconds.
Take a look at the code of the free software motion for getting inspiring ideas.
There are quite a few things to keep in mind for reliable motion detection. For example tolerate the slow changes from the sun's rotation. Or accepting momentary image glitches which can come especially from the cheapest cameras.
From a small experience I have had, I think that better than just adding up all differences, it works better to count the number of pixels whose variation exceeds a certain threshold.
Motion also offers masks, which let you for example ignore movements in a nearby road.
What about storing a black frame internally and using your same comparison code? If your new frame is different (above a threshold) from the all-black frame, start recording.
This seems the most straightforward since you already have the image-processing algorithms down.