[FFmpeg-devel] [PATCH 3/4] avdevice/decklink_dec: Added Closed caption decode from VANC
Marton Balint
cus at passwd.hu
Sun Sep 3 17:49:30 EEST 2017
On Thu, 31 Aug 2017, kjeyapal at akamai.com wrote:
> From: Karthick J <kjeyapal at akamai.com>
>
> Signed-off-by: Karthick J <kjeyapal at akamai.com>
> ---
> libavcodec/avcodec.h | 7 ++
> libavdevice/decklink_dec.cpp | 167 ++++++++++++++++++++++++++++++++++++++++---
> 2 files changed, 165 insertions(+), 9 deletions(-)
>
> diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
> index 513236a..d4a458f 100644
> --- a/libavcodec/avcodec.h
> +++ b/libavcodec/avcodec.h
> @@ -1600,6 +1600,13 @@ enum AVPacketSideDataType {
> AV_PKT_DATA_CONTENT_LIGHT_LEVEL,
>
> /**
> + * ATSC A53 Part 4 Closed Captions. This metadata should be associated with
> + * a video stream. A53 CC bitstream is stored as uint8_t in AVPacketSideData.data.
> + * The number of bytes of CC data is AVPacketSideData.size.
> + */
> + AV_PKT_DATA_A53_CC,
> +
> + /**
> * The number of side data elements (in fact a bit more than it).
> * This is not part of the public API/ABI in the sense that it may
> * change when new side data types are added.
> diff --git a/libavdevice/decklink_dec.cpp b/libavdevice/decklink_dec.cpp
> index 0b88fc8..23e1b35 100644
> --- a/libavdevice/decklink_dec.cpp
> +++ b/libavdevice/decklink_dec.cpp
> @@ -106,6 +106,13 @@ static void get_vanc_lines(int bmd_field_dominance, int height, int *field0_vanc
> }
> }
>
> +static inline unsigned parity (unsigned x)
Why don't use you simply use uint16_t here as well? I think we typically
try to avoid unsigned as a type.
> +{
> + for (unsigned i = 4 * sizeof (x); i > 0; i /= 2)
> + x ^= x >> i;
> + return x & 1;
> +}
> +
> /* The 10-bit VANC data is packed in V210, we only need the luma component. */
> static void extract_luma_from_v210(uint16_t *dst, const uint8_t *src, int width)
> {
> @@ -249,6 +256,152 @@ static uint8_t* teletext_data_unit_from_vanc_data(uint16_t *py, uint8_t *tgt, in
> return tgt;
> }
>
> +uint8_t *vanc_to_cc(AVFormatContext *avctx, uint16_t *buf, size_t words,
> + unsigned &cc_count)
> +{
> + size_t len = (buf[5] & 0xff) + 6 + 1;
> +
> + /* CDP follows */
> + uint16_t *cdp = &buf[6];
> + if (cdp[0] != 0x96 || cdp[1] != 0x69) {
> + av_log(avctx, AV_LOG_WARNING, "Invalid CDP header 0x%.2x 0x%.2x", cdp[0], cdp[1]);
> + return NULL;
> + }
> +
> + len -= 7; // remove VANC header and checksum
> +
> + if (cdp[2] != len) {
> + av_log(avctx, AV_LOG_WARNING, "CDP len %d != %zu", cdp[2], len);
> + return NULL;
> + }
> +
> + uint8_t cdp_sum = 0;
> + for (size_t i = 0; i < len - 1; i++)
> + cdp_sum += cdp[i];
> + cdp_sum = cdp_sum ? 256 - cdp_sum : 0;
> + if (cdp[len - 1] != cdp_sum) {
> + av_log(avctx, AV_LOG_WARNING, "CDP checksum invalid 0x%.4x != 0x%.4x", cdp_sum, cdp[len-1]);
> + return NULL;
> + }
> +
> + uint8_t rate = cdp[3];
> + if (!(rate & 0x0f)) {
> + av_log(avctx, AV_LOG_WARNING, "CDP frame rate invalid (0x%.2x)", rate);
> + return NULL;
> + }
> + rate >>= 4;
> + if (rate > 8) {
> + av_log(avctx, AV_LOG_WARNING, "CDP frame rate invalid (0x%.2x)", rate);
> + return NULL;
> + }
> +
> + if (!(cdp[4] & 0x43)) /* ccdata_present | caption_service_active | reserved */ {
> + av_log(avctx, AV_LOG_WARNING, "CDP flags invalid (0x%.2x)", cdp[4]);
> + return NULL;
> + }
> +
> + uint16_t hdr = (cdp[5] << 8) | cdp[6];
> + if (cdp[7] != 0x72) /* ccdata_id */ {
> + av_log(avctx, AV_LOG_WARNING, "Invalid ccdata_id 0x%.2x", cdp[7]);
> + return NULL;
> + }
> +
> + cc_count = cdp[8];
> + if (!(cc_count & 0xe0)) {
> + av_log(avctx, AV_LOG_WARNING, "Invalid cc_count 0x%.2x", cc_count);
> + return NULL;
> + }
> +
> + cc_count &= 0x1f;
> + if ((len - 13) < cc_count * 3) {
> + av_log(avctx, AV_LOG_WARNING, "Invalid cc_count %d (> %zu)", cc_count * 3, len - 13);
> + return NULL;
> + }
> +
> + if (cdp[len - 4] != 0x74) /* footer id */ {
> + av_log(avctx, AV_LOG_WARNING, "Invalid footer id 0x%.2x", cdp[len-4]);
> + return NULL;
> + }
> +
> + uint16_t ftr = (cdp[len - 3] << 8) | cdp[len - 2];
> + if (ftr != hdr) {
> + av_log(avctx, AV_LOG_WARNING, "Header 0x%.4x != Footer 0x%.4x", hdr, ftr);
> + return NULL;
> + }
> +
> + uint8_t *cc = (uint8_t *)av_malloc(cc_count * 3);
> +
> + for (size_t i = 0; i < cc_count; i++) {
> + cc[3*i + 0] = cdp[9 + 3*i+0] /* & 3 */;
> + cc[3*i + 1] = cdp[9 + 3*i+1];
> + cc[3*i + 2] = cdp[9 + 3*i+2];
> + }
> +
> + cc_count *= 3;
> + return cc;
> +}
> +
> +uint8_t *get_metadata(AVFormatContext *avctx, uint16_t *buf, size_t width,
> + uint8_t *tgt, size_t tgt_size, AVPacket *pkt)
> +{
> + decklink_cctx *cctx = (struct decklink_cctx *) avctx->priv_data;
> + uint16_t did = buf[3] & 0xFF; // data id
> + uint16_t sdid = buf[4] & 0xFF; // secondary data id
> +
> + static const uint8_t vanc_header[6] = { 0x00, 0x00, 0xff, 0x03, 0xff, 0x03 };
> + if (memcmp(vanc_header, buf, 3 * 2)) {
> + /* Does not start with the VANC header */
> + return tgt;
> + }
> +
> + size_t len = (buf[5] & 0xff) + 6 + 1;
> + if (len > width) {
> + av_log(avctx, AV_LOG_WARNING, "Data Count (%zu) > line length (%zu)\n",
> + len, width);
> + return tgt;
> + }
> +
> + uint16_t vanc_sum = 0;
> + for (size_t i = 3; i < len - 1; i++) {
> + uint16_t v = buf[i];
> + int np = v >> 8;
> + int p = parity(v & 0xff);
> + if ((!!p ^ !!(v & 0x100)) || (np != 1 && np != 2)) {
> + av_log(avctx, AV_LOG_WARNING, "Parity incorrect for word %zu\n", i);
> + return tgt;
> + }
> + vanc_sum += v;
> + vanc_sum &= 0x1ff;
> + buf[i] &= 0xff;
> + }
> +
> + vanc_sum |= ((~vanc_sum & 0x100) << 1);
> + if (buf[len - 1] != vanc_sum) {
> + av_log(avctx, AV_LOG_WARNING,
> + "VANC checksum incorrect: 0x%.4x != 0x%.4x\n", vanc_sum,
> + buf[len - 1]);
> + return tgt;
> + }
> +
> + if (did == 0x43 && (sdid == 0x02 || sdid == 0x03) && cctx->teletext_lines &&
> + width == 1920 && tgt_size >= 1920) {
> + tgt = teletext_data_unit_from_vanc_data(buf, tgt, cctx->teletext_lines);
> + } else if (did == 0x61 && sdid == 0x01) {
> + unsigned int data_len;
> + uint8_t *data = vanc_to_cc(avctx, buf, width, data_len);
> + if (data) {
> + uint8_t *pkt_cc = av_packet_new_side_data(pkt, AV_PKT_DATA_A53_CC, data_len);
> + memcpy(pkt_cc, data, data_len);
> + av_free(data);
> + }
> + } else {
> + av_log(avctx, AV_LOG_WARNING, "Unknown meta data DID = 0x%.2x SDID = 0x%.2x\n",
> + did, sdid);
> + }
This VANC parser does not seem right. VANC header can appear multiple
times, you seem to parse only the first DID. Also, you should simply
ignore unknown DID-s, ffmpeg will not be able to parse all types there
are, there is no need to warn.
I suggest you base your code on the existing VANC parser I used for OP47,
it can also handle VANC packets encapsulated in a VANC multipacket. Add
your DID/SDID-s and your parity checker code to the
teletext_data_unit_from_ancillary_packet function. (Obviously you can
rename the function to better reflect the new generic purpose). Maybe you
can split this patch even further: rename functions / add parity checker /
add A53CC.
> +
> + return tgt;
> +}
> +
> static void avpacket_queue_init(AVFormatContext *avctx, AVPacketQueue *q)
> {
> struct decklink_cctx *ctx = (struct decklink_cctx *)avctx->priv_data;
> @@ -537,7 +690,7 @@ HRESULT decklink_input_callback::VideoInputFrameArrived(
> videoFrame->GetHeight();
> //fprintf(stderr,"Video Frame size %d ts %d\n", pkt.size, pkt.pts);
>
> - if (!no_video && ctx->teletext_lines) {
> + if (!no_video) {
> IDeckLinkVideoFrameAncillary *vanc;
> AVPacket txt_pkt;
> uint8_t txt_buf0[3531]; // 35 * 46 bytes decoded teletext lines + 1 byte data_identifier + 1920 bytes OP47 decode buffer
> @@ -550,7 +703,8 @@ HRESULT decklink_input_callback::VideoInputFrameArrived(
> txt_buf[0] = 0x10; // data_identifier - EBU_data
> txt_buf++;
> #if CONFIG_LIBZVBI
> - if (ctx->bmd_mode == bmdModePAL && (vanc_format == bmdFormat8BitYUV || vanc_format == bmdFormat10BitYUV)) {
> + if (ctx->bmd_mode == bmdModePAL && ctx->teletext_lines &&
> + (vanc_format == bmdFormat8BitYUV || vanc_format == bmdFormat10BitYUV)) {
> av_assert0(videoFrame->GetWidth() == 720);
> for (i = 6; i < 336; i++, line_mask <<= 1) {
> uint8_t *buf;
> @@ -575,13 +729,8 @@ HRESULT decklink_input_callback::VideoInputFrameArrived(
> if (vanc->GetBufferForVerticalBlankingLine(i, (void**)&buf) == S_OK) {
> uint16_t luma_vanc[4096]; // Allocated for highest width (4K)
> extract_luma_from_v210(luma_vanc, buf, videoFrame->GetWidth());
> - if (videoFrame->GetWidth() == 1920) {
> - txt_buf = teletext_data_unit_from_vanc_data(luma_vanc, txt_buf, ctx->teletext_lines);
> - if (txt_buf - txt_buf0 > 1611) { // ensure we still have at least 1920 bytes free in the buffer
> - av_log(avctx, AV_LOG_ERROR, "Too many OP47 teletext packets.\n");
> - break;
> - }
> - }
> + txt_buf = get_metadata(avctx, luma_vanc, videoFrame->GetWidth(),
> + txt_buf, 3531 - (txt_buf - txt_buf0), &pkt);
> }
> if (i == field0_vanc_end)
> i = field1_vanc_start;
> --
> 1.9.1
>
Regards,
Marton
More information about the ffmpeg-devel
mailing list