Capture image with V4L2 on jetson TX2 - image-capture

Question

To explain my process, find below a diagram:
I am working on computed tomography scanner. I use jetson TX2 for image acquisition and pre-processing.
From the jetson, I control the turn table and the camera. The camera is the FSM-IMX304m. I need to access the raw pointer. For that, I need to control the camera using V4L2 (we advise not use libargus to access raw pointer, because it is store in the ISP and the ISP compress data .. Can you confirm it ?). My first problem is about the documentation about v4l2, I didn't find a clear documentation for the C++ API .. I need to control:
exposure time;
gain;
function to clear the buffer.
I found a sample on internet, see how V4L2 works :
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include <linux/ioctl.h>
#include <linux/types.h>
#include <linux/v4l2-common.h>
#include <linux/v4l2-controls.h>
#include <linux/videodev2.h>
#include <fcntl.h>
#include <unistd.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <string.h>
#include <fstream>
#include <string>
using namespace std;
int main() {
// 1. Open the device
int fd; // A file descriptor to the video device
fd = open("/dev/video0",O_RDWR);
if(fd < 0){
perror("Failed to open device, OPEN");
return 1;
}
// 2. Ask the device if it can capture frames
v4l2_capability capability;
if(ioctl(fd, VIDIOC_QUERYCAP, &capability) < 0){
// something went wrong... exit
perror("Failed to get device capabilities, VIDIOC_QUERYCAP");
return 1;
}
// 3. Set Image format
v4l2_format imageFormat;
imageFormat.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
imageFormat.fmt.pix.width = 1024;
imageFormat.fmt.pix.height = 1024;
imageFormat.fmt.pix.pixelformat = V4L2_PIX_FMT_MJPEG;
imageFormat.fmt.pix.field = V4L2_FIELD_NONE;
// tell the device you are using this format
if(ioctl(fd, VIDIOC_S_FMT, &imageFormat) < 0){
perror("Device could not set format, VIDIOC_S_FMT");
return 1;
}
// 4. Request Buffers from the device
v4l2_requestbuffers requestBuffer = {0};
requestBuffer.count = 1; // one request buffer
requestBuffer.type = V4L2_BUF_TYPE_VIDEO_CAPTURE; // request a buffer which we can use for capturing frames
requestBuffer.memory = V4L2_MEMORY_MMAP;
if(ioctl(fd, VIDIOC_REQBUFS, &requestBuffer) < 0){
perror("Could not request buffer from device, VIDIOC_REQBUFS");
return 1;
}
// 5. Query the buffer to get raw data ie. ask for the you requested buffer
// and allocate memory for it
v4l2_buffer queryBuffer = {0};
queryBuffer.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
queryBuffer.memory = V4L2_MEMORY_MMAP;
queryBuffer.index = 0;
if(ioctl(fd, VIDIOC_QUERYBUF, &queryBuffer) < 0){
perror("Device did not return the buffer information, VIDIOC_QUERYBUF");
return 1;
}
// use a pointer to point to the newly created buffer
// mmap() will map the memory address of the device to
// an address in memory
char* buffer = (char*)mmap(NULL, queryBuffer.length, PROT_READ | PROT_WRITE, MAP_SHARED,
fd, queryBuffer.m.offset);
memset(buffer, 0, queryBuffer.length);
// 6. Get a frame
// Create a new buffer type so the device knows which buffer we are talking about
v4l2_buffer bufferinfo;
memset(&bufferinfo, 0, sizeof(bufferinfo));
bufferinfo.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
bufferinfo.memory = V4L2_MEMORY_MMAP;
bufferinfo.index = 0;
// Activate streaming
int type = bufferinfo.type;
if(ioctl(fd, VIDIOC_STREAMON, &type) < 0){
perror("Could not start streaming, VIDIOC_STREAMON");
return 1;
}
/***************************** Begin looping here *********************/
// Queue the buffer
if(ioctl(fd, VIDIOC_QBUF, &bufferinfo) < 0){
perror("Could not queue buffer, VIDIOC_QBUF");
return 1;
}
// Dequeue the buffer
if(ioctl(fd, VIDIOC_DQBUF, &bufferinfo) < 0){
perror("Could not dequeue the buffer, VIDIOC_DQBUF");
return 1;
}
// Frames get written after dequeuing the buffer
cout << "Buffer has: " << (double)bufferinfo.bytesused / 1024
<< " KBytes of data" << endl;
// Write the data out to file
ofstream outFile;
outFile.open("webcam_output.jpeg", ios::binary| ios::app);
int bufPos = 0, outFileMemBlockSize = 0; // the position in the buffer and the amount to copy from
// the buffer
int remainingBufferSize = bufferinfo.bytesused; // the remaining buffer size, is decremented by
// memBlockSize amount on each loop so we do not overwrite the buffer
char* outFileMemBlock = NULL; // a pointer to a new memory block
int itr = 0; // counts thenumber of iterations
while(remainingBufferSize > 0) {
bufPos += outFileMemBlockSize; // increment the buffer pointer on each loop
// initialise bufPos before outFileMemBlockSize so we can start
// at the beginning of the buffer
outFileMemBlockSize = 1024; // set the output block size to a preferable size. 1024 :)
outFileMemBlock = new char[sizeof(char) * outFileMemBlockSize];
// copy 1024 bytes of data starting from buffer+bufPos
memcpy(outFileMemBlock, buffer+bufPos, outFileMemBlockSize);
outFile.write(outFileMemBlock,outFileMemBlockSize);
// calculate the amount of memory left to read
// if the memory block size is greater than the remaining
// amount of data we have to copy
if(outFileMemBlockSize > remainingBufferSize)
outFileMemBlockSize = remainingBufferSize;
// subtract the amount of data we have to copy
// from the remaining buffer size
remainingBufferSize -= outFileMemBlockSize;
// display the remaining buffer size
cout << itr++ << " Remaining bytes: "<< remainingBufferSize << endl;
delete outFileMemBlock;
}
// Close the file
outFile.close();
/******************************** end looping here **********************/
// end streaming
if(ioctl(fd, VIDIOC_STREAMOFF, &type) < 0){
perror("Could not end streaming, VIDIOC_STREAMOFF");
return 1;
}
close(fd);
return 0;
}
On the jetson, the code compile perfectly, but I can't run the code. It is blocked at this step :
// Dequeue the buffer
if(ioctl(fd, VIDIOC_DQBUF, &bufferinfo) < 0){
perror("Could not dequeue the buffer, VIDIOC_DQBUF");
return 1;
}
It is like the code is blocked in an endless loop. I have tested the code on my personal computer which runs Ubuntu 18.04, and the sample works well.

Answer 1

I do not have this sensor, but I assume that:
Your pixel format is incorrectly set imageFormat.fmt.pix.pixelformat = V4L2_PIX_FMT_MJPEG;
Most likely there should be 12-bit raw data from this sensor V4L2_PIX_FMT_Y12 or one of these options (Mono8/10/12/16, Bayer8/10/12/16, RGB8, YUV422, YUV411).
You can view the available formats in the Linux kernel here https://elixir.bootlin.com/linux/v4.9.237/source/include/uapi/linux/videodev2.h#L499
Check the documentation for your sensor.
Since Nvidia developers have extended the v4l2 subsystem, you need to use the following controls to adjust exposure and gain: TEGRA_CAMERA_CID_EXPOSURE, TEGRA_CAMERA_CID_GAIN. See file tegra-v4l2-camera.h
And also check the sensor controls:
v4l2-ctl --list-ctrls
....
gain 0x009a2009 (int64): min = 0 max = 480 step = 1 default = 0 value = 0 flags = slider
exposure 0x009a200a (int64): min = 28 max = 1000000 step = 1 default = 27879 value = 28 flags = slider
.....
Also examples of receiving raw data from the camera can be seen in examples from Nvidia
https://docs.nvidia.com/jetson/l4t-multimedia/mmapi_build.html