Hi there. I have begun playing around with using a wearable camera that just clips into your shirt, like w mic clip. It records MJPEG at 720x480 which is enough for my uses, recording daily walks around the city.
But there's one issue I had not taken into account: the body (actually torso) where the cam is attached to moves a LOT while walking, you might not perceive it since your neck/head does a lot of stabilization on its own, coupled with any built-in image stabilization done by the brain (great firmware!). I was shocked when I first saw the recording of me walking very fast, it looked as if was recoded from a helmet cam by Godzilla rampaging throughout the city. The problem is not only the shacking but the TILT. The body tilts a bit with each step. I have read here about the "deshake" option in FFMPEG http://adaptivesamples.com/2014/05/30/camera-stabilisation-with-ffmpeg/ But I'm curious as to what algorithm it uses, or if it is tweakable at all to compensate for side-tilt while walking in addition to vertical shake. Someone on the above post talks about the vidstabdetect and vidstabtransform filters, but without giving further detail on why those are better. Looking around, I came across this IEEE research paper (sorry,I could only read the abstract, don´t have access to it)... "A physiological camera shake model for image stabilization systems" That seems to address specifically the complexities of stabilization needed due to walking. ¨ http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6127172&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D6127172 " It successfully reproduces subjective observations made about the shape and time evolution of hand tremor-induced motion blur patterns. Our model is compared to the simpler random-walk and straight-line-walk models as well as previously reported experiments. It paves the way towards accurate deblurring and optimal image stabilization algorithms by providing realistic motion blur point spread function matrices. To our best knowledge it is the first model in which both spectral and statistical characteristics fit several ground truths from literature." In short: I´d like to know if there´s any FOSS filters -both for FFMPEG and third party ones for ie Virtualdub or Avidemux, that represents the state-of-the-art in image stabilization post-processing wrt to walking recordings... Thanks in advance! FC PS: I know what some of you will say: Just wear the camera above your head, not below your neck. Well, that's not an option for me. PS2: I´m crossposting on the dev list in case any devs are likely more aware then users... -- During times of Universal Deceit, telling the truth becomes a revolutionary act Durante épocas de Engaño Universal, decir la verdad se convierte en un Acto Revolucionario - George Orwell _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
