I've been looking at buying a stereo camera to experiment with CV, and been wondering what the best, cheap hardware is. Is there a better suited alternative than the original kinect? The xbox one version has been rising in price it seems, and the original might be too outdated these days.
I can get the original xbox 360 kinext for ~10€, the xbox one kinect has been rising in price between 30-50€. Is that still the go-to?
Hi, dear Redditors. I'm writing my thesis on using a stereovision for collision avoidance. My assumption was that collision avoidance is possible on a low power computer (RaspberryPi) using OpenCV (BM correspondence - SGBM is too slow). BM more-or-less gives me results that are OK for detecting the collision by dividing the points of cloud into segments and checking which one is free to navigate. I got a challenge to perform optimization in the process. From my explorations, the most time-consuming process is the correspondence process. I was looking for some papers around it, e.g., first hit https://www.researchgate.net/publication/224348039_A_fast_block_matching_algorthim_for_stereo_correspondence, Do you have any recommendations?
I've been playing for 20 years now. I think I've got good shots around the court. However I think my game has plateaued. One of the things I attribute this to is that I haven't got stereo vision (I.e. visual depth perception). I think its the equivalent of playing with one eye closed. I think I may to have concentrate on timing more than other people. It seems to take more evenings of squash to get into form with timing and touch. Is this reasoning likely to be a factor? Thank you.
I was wondering if anyone could recommend a good introductory textbook to stereo vision. I'm looking to make a robot that moves relative to fixed beacons, openCV will be used to facilitate programming. Thanks in advance.
Is the fundamental matrix just used to generate the epipolar line in an image plane given a pixel coordinate in the other image plane, which is useful because we just have to search along this epipolar line for the matching point of interest (in the image plane with the epipolar line)?
And just to clarify by matching points of interest, I mean the image projections gotten from the same object.
There are many different ways to collect 3D RGB-D clouds. Some sensors, like the Kinect, gives perfect data in an artificial (indoors) environments. Outdoors most of them aren’t meant to work, and they really don’t.
So I’m thinking it makes sense to limit this discussion to proven stereo vision solutions. There are many, and they work really, really well. But it’s rarely plug and play, and I’m piss poor, so I need some feedback wrt what your favorite stereo camera solution is.
I really liked ZED , until I connected it to 19.0V. But the thing that blew me away, is the Parrot SlamDunk! With some pass through and external localization, it actually creates low res clouds with an effective range of 10-20 meters outside. That’s pretty badass for a K1 embedded solution made for the consumer segment. I figure they mismatched the threshold for utilizing it with, with their target customers. But it’s really an impressive feat of engineering, at least considering how many years it’s been. Hat off.
I haven’t discovered new hardware since 2018, so forgive me for my ignorance in recent developments. Have there noen anything noteworthy? I’m using it for autonomy in outdoor unknown environments.
For context, I’m using a m600 with a LIDAR (heavy as fuck). The stereo camera will be mounted to actuators to either be used in a lighter no-lidar pixhawk attempt, or in combination with the tilt compensate mounted VLP16
Bonus Q: Is DJI M2/300 the only sensible solution? (Please say no)
I’m looking for something with similar attributes as the slamdunk, but higher end. Im especially thinking about the rate/range/resolution distribution. It doesn’t have to be as simple
I have the opportunity to take a graduate class in Computer Vision. The professor is recommending that we use Matlab. I'm worried that this is too antiquated and I need to be using Python and/or C++ to be relevant in industry. My goal is to transition to AR/VR related software roles so I thought the class might be useful. These classes take me 20 hours a week on average though so it's a big commitment with my fulltime job. Is it worth taking if we're forced to use Matlab?
Do you think there is a market for building off the shelf computer vision datasets? If so what areas are the most in need?
I went to omscentral and I’m kind of in awe of how quickly the reviews for computer vision have gone down just on the last 2 semesters. It’s a class I want to take but I don’t want to shoot myself in the foot. A lot of the students blame it on the TAs. People who took it last semester: what’s your experience? In general, do TAs roll over from semester to semester?
I subscribe to a backing track site where the sheet music plays on the screen while the track plays. I play along on trumpet.
I want to record myself and the track. I have an audio interface.
What I want to know is if I record the trumpet through the interface into one track is there a way to take the backing track directly from the site/app and record in into a second track as it plays?
Or am I best just to run an audio Jack out of the laptop into the interface - then do the trumpet separate?
Or…is there a ‘proper’/better way?
Can you note me some example where multi-agent RL framework is used to solve computer vision tasks like object detection, event detection, action recognition, tracking and so on. I am actually interested to know how multiple agents communicate/cooperate to solve a task. How the communication/ cooperation can be modeled.
