Merge pull request #250 from kerberos-io/fix/add-avc-description-fallback

fix/add-avc-description-fallback

machinery/src/capture/main.go

@@ -159,6 +159,19 @@ func HandleRecordStream(queue *packets.Queue, configDirectory string, configurat
 					}
 
 					// Close mp4
+					if len(mp4Video.SPSNALUs) == 0 && len(configuration.Config.Capture.IPCamera.SPSNALUs) > 0 {
+						mp4Video.SPSNALUs = configuration.Config.Capture.IPCamera.SPSNALUs
+					}
+					if len(mp4Video.PPSNALUs) == 0 && len(configuration.Config.Capture.IPCamera.PPSNALUs) > 0 {
+						mp4Video.PPSNALUs = configuration.Config.Capture.IPCamera.PPSNALUs
+					}
+					if len(mp4Video.VPSNALUs) == 0 && len(configuration.Config.Capture.IPCamera.VPSNALUs) > 0 {
+						mp4Video.VPSNALUs = configuration.Config.Capture.IPCamera.VPSNALUs
+					}
+					if (videoCodec == "H264" && (len(mp4Video.SPSNALUs) == 0 || len(mp4Video.PPSNALUs) == 0)) ||
+						(videoCodec == "H265" && (len(mp4Video.VPSNALUs) == 0 || len(mp4Video.SPSNALUs) == 0 || len(mp4Video.PPSNALUs) == 0)) {
+						log.Log.Warning("capture.main.HandleRecordStream(continuous): closing MP4 without full parameter sets, moov may be incomplete")
+					}
 					mp4Video.Close(&config)
 					log.Log.Info("capture.main.HandleRecordStream(continuous): recording finished: file save: " + name)
 
@@ -580,6 +593,19 @@ func HandleRecordStream(queue *packets.Queue, configDirectory string, configurat
 				lastRecordingTime = pkt.CurrentTime
 
 				// This will close the recording and write the last packet.
+				if len(mp4Video.SPSNALUs) == 0 && len(configuration.Config.Capture.IPCamera.SPSNALUs) > 0 {
+					mp4Video.SPSNALUs = configuration.Config.Capture.IPCamera.SPSNALUs
+				}
+				if len(mp4Video.PPSNALUs) == 0 && len(configuration.Config.Capture.IPCamera.PPSNALUs) > 0 {
+					mp4Video.PPSNALUs = configuration.Config.Capture.IPCamera.PPSNALUs
+				}
+				if len(mp4Video.VPSNALUs) == 0 && len(configuration.Config.Capture.IPCamera.VPSNALUs) > 0 {
+					mp4Video.VPSNALUs = configuration.Config.Capture.IPCamera.VPSNALUs
+				}
+				if (videoCodec == "H264" && (len(mp4Video.SPSNALUs) == 0 || len(mp4Video.PPSNALUs) == 0)) ||
+					(videoCodec == "H265" && (len(mp4Video.VPSNALUs) == 0 || len(mp4Video.SPSNALUs) == 0 || len(mp4Video.PPSNALUs) == 0)) {
+					log.Log.Warning("capture.main.HandleRecordStream(motiondetection): closing MP4 without full parameter sets, moov may be incomplete")
+				}
 				mp4Video.Close(&config)
 				log.Log.Info("capture.main.HandleRecordStream(motiondetection): file save: " + name)
 

machinery/src/video/mp4.go

@@ -13,6 +13,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/Eyevinn/mp4ff/avc"
 	mp4ff "github.com/Eyevinn/mp4ff/mp4"
 	"github.com/kerberos-io/agent/machinery/src/encryption"
 	"github.com/kerberos-io/agent/machinery/src/log"
@@ -158,6 +159,68 @@ func (mp4 *MP4) AddAudioTrack(codec string) uint32 {
 func (mp4 *MP4) AddMediaSegment(segNr int) {
 }
 
+// updateVideoParameterSetsFromAnnexB inspects Annex B data to fill missing SPS/PPS/VPS.
+func (mp4 *MP4) updateVideoParameterSetsFromAnnexB(data []byte) {
+	if len(data) == 0 {
+		return
+	}
+
+	needSPS := len(mp4.SPSNALUs) == 0
+	needPPS := len(mp4.PPSNALUs) == 0
+	needVPS := len(mp4.VPSNALUs) == 0
+	if !(needSPS || needPPS || needVPS) {
+		return
+	}
+
+	for _, nalu := range splitNALUs(data) {
+		nalu = removeAnnexBStartCode(nalu)
+		if len(nalu) == 0 {
+			continue
+		}
+
+		switch mp4.VideoTrackName {
+		case "H264", "AVC1":
+			nalType := nalu[0] & 0x1F
+			switch nalType {
+			case 7: // SPS
+				if needSPS {
+					mp4.SPSNALUs = [][]byte{nalu}
+					needSPS = false
+					log.Log.Warning("mp4.updateVideoParameterSetsFromAnnexB(): SPS recovered from in-band NALU")
+				}
+			case 8: // PPS
+				if needPPS {
+					mp4.PPSNALUs = [][]byte{nalu}
+					needPPS = false
+					log.Log.Warning("mp4.updateVideoParameterSetsFromAnnexB(): PPS recovered from in-band NALU")
+				}
+			}
+		case "H265", "HVC1":
+			nalType := (nalu[0] >> 1) & 0x3F
+			switch nalType {
+			case 32: // VPS
+				if needVPS {
+					mp4.VPSNALUs = [][]byte{nalu}
+					needVPS = false
+					log.Log.Warning("mp4.updateVideoParameterSetsFromAnnexB(): VPS recovered from in-band NALU")
+				}
+			case 33: // SPS
+				if needSPS {
+					mp4.SPSNALUs = [][]byte{nalu}
+					needSPS = false
+					log.Log.Warning("mp4.updateVideoParameterSetsFromAnnexB(): SPS recovered from in-band NALU")
+				}
+			case 34: // PPS
+				if needPPS {
+					mp4.PPSNALUs = [][]byte{nalu}
+					needPPS = false
+					log.Log.Warning("mp4.updateVideoParameterSetsFromAnnexB(): PPS recovered from in-band NALU")
+				}
+			}
+		}
+	}
+}
+
 // flushPendingVideoSample writes the pending video sample to the current fragment.
 // If nextPTS is provided (non-zero), it calculates duration from the PTS difference.
 // If nextPTS is 0 (e.g., at Close time), it uses the last known duration.
@@ -283,6 +346,7 @@ func (mp4 *MP4) AddSampleToTrack(trackID uint32, isKeyframe bool, data []byte, p
 	if mp4.Start {
 
 		if trackID == uint32(mp4.VideoTrack) {
+			mp4.updateVideoParameterSetsFromAnnexB(data)
 
 			var lengthPrefixed []byte
 			var err error
@@ -368,7 +432,12 @@ func (mp4 *MP4) Close(config *models.Config) {
 		mp4.TotalKeyframesReceived, mp4.TotalKeyframesWritten, mp4.SegmentCount, mp4.FragmentKeyframeCount))
 
 	if mp4.VideoTotalDuration == 0 && mp4.AudioTotalDuration == 0 {
-		log.Log.Error("mp4.Close(): no video or audio samples added, cannot create MP4 file")
+		log.Log.Error("mp4.Close(): no video or audio samples added, removing empty MP4 file")
+		mp4.Writer.Flush()
+		_ = mp4.FileWriter.Sync()
+		_ = mp4.FileWriter.Close()
+		_ = os.Remove(mp4.FileName)
+		return
 	}
 
 	// Add final pending samples before closing
@@ -491,8 +560,16 @@ func (mp4 *MP4) Close(config *models.Config) {
 	case "H264", "AVC1":
 		init.AddEmptyTrack(videoTimescale, "video", "und")
 		includePS := true
-		err := init.Moov.Traks[0].SetAVCDescriptor("avc1", mp4.SPSNALUs, mp4.PPSNALUs, includePS)
+		spsNALUs, ppsNALUs := normalizeH264ParameterSets(mp4.SPSNALUs, mp4.PPSNALUs)
+		log.Log.Debug("mp4.Close(): AVC parameter sets: SPS=" + formatNaluDebug(spsNALUs) + ", PPS=" + formatNaluDebug(ppsNALUs))
+		err := init.Moov.Traks[0].SetAVCDescriptor("avc1", spsNALUs, ppsNALUs, includePS)
 		if err != nil {
+			log.Log.Error("mp4.Close(): error setting AVC descriptor: " + err.Error())
+			if fallbackErr := addAVCDescriptorFallback(init.Moov.Traks[0], spsNALUs, ppsNALUs, uint16(mp4.width), uint16(mp4.height)); fallbackErr != nil {
+				log.Log.Error("mp4.Close(): error setting AVC descriptor fallback: " + fallbackErr.Error())
+			} else {
+				log.Log.Warning("mp4.Close(): AVC descriptor fallback used due to SPS parse error")
+			}
 		}
 		init.Moov.Traks[0].Tkhd.Duration = actualVideoDuration
 		init.Moov.Traks[0].Tkhd.Width = mp4ff.Fixed32(uint32(mp4.width) << 16)
@@ -509,8 +586,11 @@ func (mp4 *MP4) Close(config *models.Config) {
 	case "H265", "HVC1":
 		init.AddEmptyTrack(videoTimescale, "video", "und")
 		includePS := true
-		err := init.Moov.Traks[0].SetHEVCDescriptor("hvc1", mp4.VPSNALUs, mp4.SPSNALUs, mp4.PPSNALUs, [][]byte{}, includePS)
+		vpsNALUs, spsNALUs, ppsNALUs := normalizeH265ParameterSets(mp4.VPSNALUs, mp4.SPSNALUs, mp4.PPSNALUs)
+		log.Log.Debug("mp4.Close(): HEVC parameter sets: VPS=" + formatNaluDebug(vpsNALUs) + ", SPS=" + formatNaluDebug(spsNALUs) + ", PPS=" + formatNaluDebug(ppsNALUs))
+		err := init.Moov.Traks[0].SetHEVCDescriptor("hvc1", vpsNALUs, spsNALUs, ppsNALUs, [][]byte{}, includePS)
 		if err != nil {
+			log.Log.Error("mp4.Close(): error setting HEVC descriptor: " + err.Error())
 		}
 		init.Moov.Traks[0].Tkhd.Duration = actualVideoDuration
 		init.Moov.Traks[0].Tkhd.Width = mp4ff.Fixed32(uint32(mp4.width) << 16)
@@ -524,8 +604,8 @@ func (mp4 *MP4) Close(config *models.Config) {
 		init.Moov.Traks[0].Mdia.Mdhd.ModificationTime = macTime
 	}
 
-	// Try adding audio track if available
-	if mp4.AudioTrackName == "AAC" || mp4.AudioTrackName == "MP4A" {
+	// Try adding audio track if available and samples were recorded.
+	if (mp4.AudioTrackName == "AAC" || mp4.AudioTrackName == "MP4A") && mp4.AudioTotalDuration > 0 {
 		// Add an audio track to the moov box
 		init.AddEmptyTrack(audioTimescale, "audio", "und")
 
@@ -763,6 +843,172 @@ func removeAnnexBStartCode(nalu []byte) []byte {
 	return nalu
 }
 
+// sanitizeParameterSets removes Annex B start codes and drops empty NALUs.
+func sanitizeParameterSets(nalus [][]byte) [][]byte {
+	if len(nalus) == 0 {
+		return nalus
+	}
+	clean := make([][]byte, 0, len(nalus))
+	for _, nalu := range nalus {
+		trimmed := removeAnnexBStartCode(nalu)
+		if len(trimmed) == 0 {
+			continue
+		}
+		clean = append(clean, trimmed)
+	}
+	return clean
+}
+
+// normalizeH264ParameterSets splits Annex B blobs and extracts SPS/PPS NALUs.
+func normalizeH264ParameterSets(spsIn [][]byte, ppsIn [][]byte) ([][]byte, [][]byte) {
+	all := make([][]byte, 0, len(spsIn)+len(ppsIn))
+	all = append(all, spsIn...)
+	all = append(all, ppsIn...)
+	var spsOut [][]byte
+	var ppsOut [][]byte
+	for _, blob := range all {
+		for _, nalu := range splitParamSetNALUs(blob) {
+			nalu = removeAnnexBStartCode(nalu)
+			if len(nalu) == 0 {
+				continue
+			}
+			typ := nalu[0] & 0x1F
+			switch typ {
+			case 7:
+				spsOut = append(spsOut, nalu)
+			case 8:
+				ppsOut = append(ppsOut, nalu)
+			}
+		}
+	}
+	if len(spsOut) == 0 {
+		spsOut = sanitizeParameterSets(spsIn)
+	}
+	if len(ppsOut) == 0 {
+		ppsOut = sanitizeParameterSets(ppsIn)
+	}
+	return spsOut, ppsOut
+}
+
+// normalizeH265ParameterSets splits Annex B blobs and extracts VPS/SPS/PPS NALUs.
+func normalizeH265ParameterSets(vpsIn [][]byte, spsIn [][]byte, ppsIn [][]byte) ([][]byte, [][]byte, [][]byte) {
+	all := make([][]byte, 0, len(vpsIn)+len(spsIn)+len(ppsIn))
+	all = append(all, vpsIn...)
+	all = append(all, spsIn...)
+	all = append(all, ppsIn...)
+	var vpsOut [][]byte
+	var spsOut [][]byte
+	var ppsOut [][]byte
+	for _, blob := range all {
+		for _, nalu := range splitParamSetNALUs(blob) {
+			nalu = removeAnnexBStartCode(nalu)
+			if len(nalu) == 0 {
+				continue
+			}
+			typ := (nalu[0] >> 1) & 0x3F
+			switch typ {
+			case 32:
+				vpsOut = append(vpsOut, nalu)
+			case 33:
+				spsOut = append(spsOut, nalu)
+			case 34:
+				ppsOut = append(ppsOut, nalu)
+			}
+		}
+	}
+	if len(vpsOut) == 0 {
+		vpsOut = sanitizeParameterSets(vpsIn)
+	}
+	if len(spsOut) == 0 {
+		spsOut = sanitizeParameterSets(spsIn)
+	}
+	if len(ppsOut) == 0 {
+		ppsOut = sanitizeParameterSets(ppsIn)
+	}
+	return vpsOut, spsOut, ppsOut
+}
+
+// splitParamSetNALUs splits Annex B parameter set blobs; raw NALUs are returned as-is.
+func splitParamSetNALUs(blob []byte) [][]byte {
+	if len(blob) == 0 {
+		return nil
+	}
+	if findStartCode(blob, 0) >= 0 {
+		return splitNALUs(blob)
+	}
+	return [][]byte{blob}
+}
+
+func formatNaluDebug(nalus [][]byte) string {
+	if len(nalus) == 0 {
+		return "none"
+	}
+	parts := make([]string, 0, len(nalus))
+	for _, nalu := range nalus {
+		if len(nalu) == 0 {
+			parts = append(parts, "len=0")
+			continue
+		}
+		max := 8
+		if len(nalu) < max {
+			max = len(nalu)
+		}
+		parts = append(parts, fmt.Sprintf("len=%d head=%x", len(nalu), nalu[:max]))
+	}
+	return strings.Join(parts, "; ")
+}
+
+func addAVCDescriptorFallback(trak *mp4ff.TrakBox, spsNALUs, ppsNALUs [][]byte, width, height uint16) error {
+	if trak == nil || trak.Mdia == nil || trak.Mdia.Minf == nil || trak.Mdia.Minf.Stbl == nil || trak.Mdia.Minf.Stbl.Stsd == nil {
+		return fmt.Errorf("missing trak stsd")
+	}
+	if len(spsNALUs) == 0 {
+		return fmt.Errorf("no SPS NALU available")
+	}
+	decConfRec, err := buildAVCDecConfRecFromSPS(spsNALUs, ppsNALUs)
+	if err != nil {
+		return err
+	}
+	if width == 0 && trak.Tkhd != nil {
+		width = uint16(uint32(trak.Tkhd.Width) >> 16)
+	}
+	if height == 0 && trak.Tkhd != nil {
+		height = uint16(uint32(trak.Tkhd.Height) >> 16)
+	}
+	if width > 0 && height > 0 && trak.Tkhd != nil {
+		trak.Tkhd.Width = mp4ff.Fixed32(uint32(width) << 16)
+		trak.Tkhd.Height = mp4ff.Fixed32(uint32(height) << 16)
+	}
+	avcC := &mp4ff.AvcCBox{DecConfRec: *decConfRec}
+	avcx := mp4ff.CreateVisualSampleEntryBox("avc1", width, height, avcC)
+	trak.Mdia.Minf.Stbl.Stsd.AddChild(avcx)
+	return nil
+}
+
+func buildAVCDecConfRecFromSPS(spsNALUs, ppsNALUs [][]byte) (*avc.DecConfRec, error) {
+	if len(spsNALUs) == 0 {
+		return nil, fmt.Errorf("no SPS NALU available")
+	}
+	sps := spsNALUs[0]
+	if len(sps) < 4 {
+		return nil, fmt.Errorf("SPS too short: len=%d", len(sps))
+	}
+	// SPS NALU: byte 0 is NAL header, next 3 bytes are profile/compat/level.
+	dec := &avc.DecConfRec{
+		AVCProfileIndication: sps[1],
+		ProfileCompatibility: sps[2],
+		AVCLevelIndication:   sps[3],
+		SPSnalus:             spsNALUs,
+		PPSnalus:             ppsNALUs,
+		ChromaFormat:         1,
+		BitDepthLumaMinus1:   0,
+		BitDepthChromaMinus1: 0,
+		NumSPSExt:            0,
+		NoTrailingInfo:       true,
+	}
+	return dec, nil
+}
+
 // splitNALUs splits Annex B data into raw NAL units without start codes.
 func splitNALUs(data []byte) [][]byte {
 	var nalus [][]byte