I am using a 160x120 pixel IR camera from FLiR (the Lepton 3.5), and I want to detect face with it so I can get temperature data from a person's face. However, I am having a hard time finding a pre-trained program for face detection in infrared images. Anyone who have any tips? If not, anyone who have successfully trained a haar cascade (or another algorithm) to detect objects/faces in IR images?
Super grateful for any help :)
Cheers
Have you tried normalizing your images so that they become grayscale images in the range of 0 and 255? OpenCV's haarcascade_frontalface_alt2.xml must work on the normalized IR images.
Related
I am collecting data for a project. The data collection is done by recording videos of the subjects and the environment. However, while training the network, I would not want to train it with all the images collected in the video sequence.
The main objective is to not train the network with redundant images. The video sequence collected at 30 frames/sec can have redundant images (images that are very similar) within the short intervals. T(th) frame and (T+1)th frame can be similar.
Can someone suggest ways to extract only the images that can be useful for training ?
Update #2: Further resources,
https://github.com/JohannesBuchner/imagehash
https://www.pyimagesearch.com/2017/11/27/image-hashing-opencv-python/
https://www.pyimagesearch.com/2020/04/20/detect-and-remove-duplicate-images-from-a-dataset-for-deep-learning/
Update #1: You can use this repo to calculate similarity between given images. https://github.com/quickgrid/image-similarity**
If frames with certain objects(e.g., vehicle, device) are important, then use pretrained object detectors if available, to extract important frames.
Next, use a similarity method to remove similar images in nearby frames. Until a chosen threshold is exceeded keep removing nearby N frames.
This link should be helpful in finding right method for your case,
https://datascience.stackexchange.com/questions/48642/how-to-measure-the-similarity-between-two-images
This repository below should help implement the idea with few lines of code. It uses CNN to extract features then calculates there cosine distance as mentioned there.
https://github.com/ryanfwy/image-similarity
Till now I have been able to create an application where the Kinect sensor is at one place. I have used speech recognition EmguCV (open cv) and Aforge.NET to help me process an image, learn and recognize objects. It all works fine but there is always scope for improvement and I am posing some problems: [Ignore the first three I want the answer for the fourth]
The frame rate is horrible. Its like 5 fps even though it should be like 30 fps. (This is WITHOUT all the processing) My application is running fine, it gets color as well as depth frames from the camera and displays it. Still the frame rate is bad. The samples run awesome, around 25 fps. Even though I ran the exact same code from the samples it wont just budge. :-( [There is no need for code, please tell me the possible problems.]
I would like to create a little robot on which the kinect and my laptop will be mounted on. I tried using the Mindstorms Kit but the lowtorque motors dont do the trick. Please tell me how will I achieve this.
How do I supply power on board? I know that the Kinect uses 12 volts for the motor. But it gets that from an AC adapter. [I would not like to cut my cable and replace it with a 12 volt battery]
The biggest question: How in this world will it navigate. I have done A* and flood-fill algorithms. I read this paper like a thousand times and I got nothing. I have the navigation algorithm in my mind but how on earth will it localize itself? [It should not use GPS or any kind of other sensors, just its eyes i.e. the Kinect]
Helping me will be Awesome. I am a newbie so please don't expect me to know everything. I have been up on the internet for 2 weeks with no luck.
Thanks A lot!
Localisation is a tricky task, as it depends on having prior knowledge of the environment in which your robot will be placed (i.e. a map of your house). While algorithms exist for simultaneous localisation and mapping, they tend to be domain-specific and as such not applicable to the general case of placing a robot in an arbitrary location and having it map its environment autonomously.
However, if your robot does have a rough (probabilistic) idea of what its environment looks like, Monte Carlo localisation is a good choice. On a high level, it goes something like:
Firstly, the robot should make a large number of random guesses (called particles) as to where it could possibly be within its known environment.
With each update from the sensor (i.e. after the robot has moved a short distance), it adjusts the probability that each of its random guesses is correct using a statistical model of its current sensor data. This can work especially well if the robot takes 360ยบ sensor measurements, but this is not completely necessary.
This lecture by Andrew Davison at Imperial College London gives a good overview of the mathematics involved. (The rest of the course will most likely be very interesting to you as well, given what you are trying to create). Good luck!
I am trying to record a stream from a webcam using Expression Encoder 4 SDK in WPF I can capture the video & audio streams and record these to disk however they are only recording at a base resolution of 320x240 the webcam is capable of capturing at 720p, how can I record at this resolution. Any help would be appreciated, I have been pulling my hair out trying to solve this all week.
Know this is a bit late but all questions need answers:
These might be a possible solution:
Check to see if your camera has it's own settings on the camera or comes with an installation disk.
for the expression encoder 4 put the video profile quality to max.
Good luck. If you are still around tell me, how it goes.
to change the "size" you can use the following line :
LiveJob.OutputFormat.VideoProfile.Streams[0].Size = new Size(1280,1080)
Or whatever you want it to be.
Encoder also offers a setting page that you can use.
That's what I did and then after setting the outputsize you can do that :
currentJob.OutputFormat.VideoProfile.Streams[0].Size = ((LiveSource)LiveDeviceSource).CropRect.Size;
Only 1 small limitation, you can't change the size while it's recording if you are publishing the source.
I am working for a project at school regarding face detection, based on a technique described by Viola and Jones 2001/2004.
I've read that the OpenCV has an implementation of this algorithm, and it works very good.
I was wondering if you have any advices regarding what techniques (pre-processing) to apply to the images before testing the existence of a face (eg. histogram equalization) ?
I basically used the code from this sample program from the OpenCV page and it worked very well for my masters thesis project. If you get bad results or your lighting is strange you can try a histogram equalization.
with a friend I did something similar too for an university project, and especially on low resolution video sequences it really helped to upsample the frame, doubling its size. It was my friends' idea, who had previously taken an image processing class. Although equivalent, things like decreasing initial scan window size, horizontal and vertical steps didn't produce the same result. In other words it may be better to work on larger images with larger scan windows than on smaller with smaller scan windows. Don't know exactly why.
Bye ;-)
I know its too late. But do go through this site as well.
It speaks of the common pre-proccessing required for the images. Equalising the image, Editing irrelevant content etc
I'm looking for the most realistic way of playing sound of a rolling ball. Currently I'm using a Wav sample that I play over and over as long as the ball is moving - which just doesn't feel right.
I've been thinking about completely synthesizing the sound, which I know very little about (almost nothing), I'd be grateful for any tutorials/research materials/samples concerning synthesis of sound of a ball made of particular material rolling on surface made of another material. Also if this idea is completely wrong, please suggest another way of doing this.
Thanks!
I would guess that you'll get the biggest bang for your buck by doing a dynamic frequency adjustment on the sound that makes the playback frequency proportional to the velocity of the ball. I don't know what type of sound library you use, but most will support some variant of this.
For example, in FMOD you could use the Channel::setFrequency method. Ideally, you would compute your desired playback frequency based on your WAV's original sample frequency (Fo), the ball's current velocity (Vc), and the ball's 'ideal' velocity at which the default WAV sounds right (Vi). Something generally like:
F = Fo * ( Vc / Vi )
This will tend to break down as the ball gets farther away from the 'ideal' velocity. You might want to have several different WAVs that are appropriate for different speed ranges that you switch to at certain threshold velocities. Within each WAV's bracket, you'd do the same kind of frequency adjustment.
Another note: this is probably not something that is worth doing every frame. I'd guess that doing this more than 20 times per second would be a waste of time.
ADDENDUM: Playback frequency scaling like this can also be used for simulating the Doppler effect as well. Once you have your adjusted playback frequency, you'd perform another scale of the Frequency based on the velocity of the ball relative to the 'listener' (the camera).
Have you tried playing the sound forward, then playing it backward, and looping that? I use this trick graphically to creating repeating patterns. I don't know much about sound but it might work?
One approach might be to analyze the sound of a rolling ball, and decompose it into its component waveforms. Then you'd be able to generate your own wav file with synthesized waves.
You should be able to do this using an FFT on a sample of the sound.
One drawback is that the sound will likely sound synthesized - you'll have to add noise and such to make it sound more realistic. Getting it to sound real enough may be the hardest part.
I don't think you need the trouble to synthesize that. It would be way too hard to even sound convincing.
Depending on how your scene is, you could loop the sound foward/backwards and simulate a doppler effect applying a low pass filter and/or changing it's pitch.
By the way, freesoung.org is a great place for free samples. They are not professionally recorded but are a good starting point for manipulation. On the other hand, sound ideas has some great sample cds (they're actually industry standard) if you can find them on the cheap. You just have to search for which one has rolling ball sounds.
I really like the approach described in this SIGGRAPH paper:
http://www.cs.ubc.ca/~kvdoel/publications/foleyautomatic.pdf
It describes synthesizing the sound of a rock rolling in a wok (no, really :). The idea is to use modal synthesis (i.e. convolved impulse responses), and the results can be very convincing.
Here's a link to the video demo that goes with the paper:
http://www.cs.ubc.ca/~kvdoel/publications/foleyautomatic.mpeg
And here's a link to the JASS library (written by one of the authors), which was used to create the sound for the video:
http://www.cs.ubc.ca/~kvdoel/jass/jass.html
I'm not sure if you could make it run on a smart phone, but with an efficient enough convolution routine/approximation you might be able to do something interesting...
My question is 'why?' - do you see some benefit in this, or is it just for fun? Your question implies that you aren't happy with the wav you are using but I strongly believe that synthesising your own is going to sound far inferior.
If your wav sample doesn't sound right, I'd suggest try to find another sample. Synthesising a sound is not easy and is never going to sound as realistic as your sample.
Real time synthesis may require more resources for processing and computation. You may very well end up prerendering your synthesised sound into a wav file and performing a playback.
If you want to simulate the sound of different materials then you can use some DSP, or even simple tricks like slowing or speeding the wav playback. The simplest way is the prerender these in another application and store one copy of the file for each use.