您好!我一直在研究机器人项目的软件,该项目将视觉作为主要感官之一。我一直在使用内核4.9.35-v7 +运行Raspbian Jessie Lite的Raspberry Pi 3构建系统。除了其他板载传感器之外,我还计划在摄像机上记录一些ML模型。要检索视频帧,请使用带有a Minoru 3D USB网络摄像头的v4l2 api。我的项目的来源可以在https://github.com/mrpossoms/AVC2017找到。
这是一个非常奇怪的问题。我上面描述的那个工作正常,除了帧速率从目标15fps显着下降,但只有当它看到超过1米远的物体时!我录制了一段显示问题的视频here。首先,我将手放在相机附近,您可以在终端中看到帧速率从理想的15fps开始。然而,当我慢慢移开手时,帧速率下降到7fps左右!
这是我的相机代码:
#include <sys/ioctl.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <string.h>
#include <strings.h>
#include <fcntl.h>
#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
#include <assert.h>
#include "structs.h"
#include "cam.h"
#define CAM_FPS 5
cam_t cam_open(const char* path, cam_settings_t* cfg)
{
int fd = open(path, O_RDWR);
int res;
if(fd < 0)
{
fprintf(stderr, "Error opening video device '%s'\n", path);
//exit(-1);
cam_t empty = {};
return empty;
}
struct v4l2_capability cap;
res = ioctl(fd, VIDIOC_QUERYCAP, &cap);
if(res < 0)
{
fprintf(stderr, "Error: %d querying '%s' for capabilities (%d)\n", res, path, errno);
exit(-2);
}
if(!(cap.capabilities & V4L2_CAP_VIDEO_CAPTURE))
{
fprintf(stderr, "Error: '%s' lacks V4L2_CAP_VIDEO_CAPTURE capability\n", path);
}
res = cam_config(fd, cfg);
if(res < 0)
{
fprintf(stderr, "Error: %d configuring '%s' (%d)\n", res, path, errno);
exit(-3);
}
// Inform v4l about the buffers we want to receive data through
struct v4l2_requestbuffers bufrequest = {};
bufrequest.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
bufrequest.memory = V4L2_MEMORY_MMAP;
bufrequest.count = 2;
if(ioctl(fd, VIDIOC_REQBUFS, &bufrequest) < 0)
{
fprintf(stderr, "VIDIOC_REQBUFS\n");
exit(-4);
}
if(bufrequest.count < 2)
{
fprintf(stderr, "Not enough memory\n");
exit(-5);
}
struct v4l2_buffer bufferinfo = {};
void** fbs = calloc(sizeof(void*), bufrequest.count);
for(int i = bufrequest.count; i--;)
{
bufferinfo.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
bufferinfo.memory = V4L2_MEMORY_MMAP;
bufferinfo.index = i;
if(ioctl(fd, VIDIOC_QUERYBUF, &bufferinfo) < 0)
{
fprintf(stderr, "VIDIOC_QUERYBUF\n");
exit(-5);
}
fbs[i] = mmap(
NULL,
bufferinfo.length,
PROT_READ | PROT_WRITE,
MAP_SHARED,
fd,
bufferinfo.m.offset
);
if(fbs[i] == MAP_FAILED)
{
fprintf(stderr, "mmap failed\n");
exit(-6);
}
bzero(fbs[i], bufferinfo.length);
ioctl(fd, VIDIOC_QBUF, &bufferinfo);
}
cam_t cam = {
.fd = fd,
.frame_buffers = fbs,
.buffer_info = bufferinfo,
};
cam_wait_frame(&cam);
int type = bufferinfo.type;
if(ioctl(fd, VIDIOC_STREAMON, &type) < 0)
{
fprintf(stderr, "Error starting streaming\n");
exit(-7);
}
return cam;
}
int cam_request_frame(cam_t* cam)
{
cam->buffer_info.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
cam->buffer_info.memory = V4L2_MEMORY_MMAP;
return ioctl(cam->fd, VIDIOC_QBUF, &cam->buffer_info);
}
int cam_wait_frame(cam_t* cam)
{
ioctl(cam->fd, VIDIOC_DQBUF, &cam->buffer_info);
}
int cam_config(int fd, cam_settings_t* cfg)
{
int res = 0;
struct v4l2_format format;
if(!cfg)
{
fprintf(stderr, "Error: null configuration provided\n");
return -1;
}
/*
res = ioctl(fd, VIDIOC_G_FMT, &format);
if(res < 0)
{
fprintf(stderr, "Error: failed retrieving camera settings (%d)\n", errno);
return -2;
}
*/
format.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
format.fmt.pix.pixelformat = V4L2_PIX_FMT_YUYV;
format.fmt.pix.width = cfg->width;
format.fmt.pix.height = cfg->height;
if(ioctl(fd, VIDIOC_S_FMT, &format) < 0)
{
fprintf(stderr, "Error: failed applying camera settings\n");
return -3;
}
struct v4l2_streamparm parm = {};
parm.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
parm.parm.capture.timeperframe.numerator = 1;
parm.parm.capture.timeperframe.denominator = 15;
res = ioctl(fd, VIDIOC_S_PARM, &parm);
return 0;
}
我可以100%的时间重现这个问题,而且我真的不知道是什么原因导致这种行为。任何建议或想法都非常感谢。
答案 0 :(得分:1)
正如马克所说,问题在于曝光。驾驶员必须确定在较暗的情况下每帧需要更多的曝光时间。只需将相机移到外面,立即将帧速率提高到目标值15fps。