I'm using V4L2 to capture images from a web cam. I've tried: - mmap() - read() - libv4l2 - without libv4l2 At low resolutions, such as 160x120 and 320x240, things are typically good. But starting at 640x480 and higher resolutions, I often get only 10% to 25% of the image. The rest of the bytes in the buffer are left untouched. When using mmap buffers, v4l2_buffer.bytesused is much less than v4l2_buffer.length. When using read, the ssize_t count is less than the full image. I'm using a native non-compressed YUYV 4:2:2 format, so I know exactly how many bytes to expect. When the full image doesn't come in, it looks like this: https://www.ccoderun.ca/tmp/partial_image.jpg In case it helps, here is some relevant V4L information I dump out after the capture. Note the bytes used and length of the memory mapped buffer, which corresponds to the image linked above: Filename: "/dev/video0" Driver name: uvcvideo v4.4.40 Device name: USB 2.0 Camera Location: usb-0000:00:0b.0-1 Flags: 0x85200001 The device supports the single-planar API through the Video Capture interface. The device supports the struct v4l2_pix_format extended fields. The device supports the read() and/or write() I/O methods. The device supports the streaming I/O method. The driver fills the device_caps field. Currently selected format: type: V4L2_BUF_TYPE_VIDEO_CAPTURE width: 1024 height: 768 pixel format: YUYV field: V4L2_FIELD_NONE bytes per line: 2048 bytes image size: 1572864 bytes colour space: V4L2_COLORSPACE_DEFAULT Memory mapped buffer #0: type: V4L2_BUF_TYPE_VIDEO_CAPTURE bytes used: 284880 bytes flags: 73733 field: V4L2_FIELD_NONE sequence: 0 length: 1572864 bytes address: 0xb7197000 I've checked things like errno and xioctl results to ensure no errors are reported. Does anyone have a guess as to why the buffers aren't always full, or what I must do to ensure I get the whole buffer/read? TIA, Stéphane Charette
