[FFmpeg-devel] [PATCH v3 2/5] avfilter/vf_libopencv: add opencv HaarCascade classifier simple face detection filter
Paul B Mahol
onemda at gmail.com
Thu May 21 17:44:26 EEST 2020
On 5/21/20, lance.lmwang at gmail.com <lance.lmwang at gmail.com> wrote:
> On Thu, May 21, 2020 at 04:12:56PM +0200, Paul B Mahol wrote:
>> On 5/18/20, lance.lmwang at gmail.com <lance.lmwang at gmail.com> wrote:
>> > On Mon, May 18, 2020 at 01:11:12PM +0200, Paul B Mahol wrote:
>> >> This opencv module is obsolete for latest opencv.
>> >>
>> >> Instead there should be C++ wrapper.
>> >
>> > Sorry, where is the C++ wrapper? Now Opencv 2.x and 3.x is good
>> > to use for my current function. By my initial testing, the performance
>> > is
>> > very good, we can easily to get realtime for HD, the drawbacks are:
>> > 1. The face is not directly opposite and cannot be recognized.
>> > 2. there are some false detect for video.
>> >
>> > If have any old discussion for that, please give the discussion link for
>> > reference. I'm not keep track of the ML before.
>>
>> Nope, i'm not doing others people work for free.
>
> So are you against continuing to develop existing opencv features？Now I
> have tested with opencv 2.10 and 3.10, 3.13, but > 4.0 will not working as
> expected for known issue.
Yes, and I want that known issue with latest opencv to be properly
fixed, instead
of using shortcut hacks that will work with old versions only.
>
>>
>> >
>> >
>> >>
>> >> On 5/18/20, lance.lmwang at gmail.com <lance.lmwang at gmail.com> wrote:
>> >> > From: Limin Wang <lance.lmwang at gmail.com>
>> >> >
>> >> > Signed-off-by: Limin Wang <lance.lmwang at gmail.com>
>> >> > ---
>> >> > change the update_metadata() to postprocess() only, I'll add opencv
>> >> > drawbox
>> >> > filter and it need preprocess() to get the meta, so I prefer to
>> >> > change
>> >> > the
>> >> > function name for better readablity, in future, it may have other
>> >> > processing
>> >> > than metadata only.
>> >> >
>> >> > configure | 1 +
>> >> > doc/filters.texi | 29 +++++++
>> >> > libavfilter/vf_libopencv.c | 164
>> >> > ++++++++++++++++++++++++++++++++++++-
>> >> > 3 files changed, 191 insertions(+), 3 deletions(-)
>> >> >
>> >> > diff --git a/configure b/configure
>> >> > index 34afdaad28..281b67efc4 100755
>> >> > --- a/configure
>> >> > +++ b/configure
>> >> > @@ -2123,6 +2123,7 @@ HEADERS_LIST="
>> >> > machine_ioctl_meteor_h
>> >> > malloc_h
>> >> > opencv2_core_core_c_h
>> >> > + opencv2_objdetect_objdetect_c_h
>> >> > OpenGL_gl3_h
>> >> > poll_h
>> >> > sys_param_h
>> >> > diff --git a/doc/filters.texi b/doc/filters.texi
>> >> > index d9ba0fffa1..f938dd04de 100644
>> >> > --- a/doc/filters.texi
>> >> > +++ b/doc/filters.texi
>> >> > @@ -14177,6 +14177,35 @@ other parameters is 0.
>> >> > These parameters correspond to the parameters assigned to the
>> >> > libopencv function @code{cvSmooth}.
>> >> >
>> >> > + at subsection facedetect
>> >> > +Face detection using Haar Feature-based Cascade Classifiers.
>> >> > +
>> >> > +The filter takes the following parameters:
>> >> > + at var{xml_model}|@var{qoffset}.
>> >> > +
>> >> > + at var{xml_model} is the path of pre-trained classifiers, The C API
>> >> > still
>> >> > +does not support the newer cascade format, please use the old format
>> >> > +haarcascade_frontalface_alt.xml which type_id is
>> >> > opencv-haar-classifier.
>> >> > +
>> >> > + at var{qoffset}
>> >> > +If you want export the detected faces by ROI side data in frame,
>> >> > please
>> >> > set
>> >> > the
>> >> > +parameters, See also the @ref{addroi} filter. The range of qoffset
>> >> > is
>> >> > from
>> >> > [-1.0, 1.0]
>> >> > +
>> >> > +By default the filter will report these metadata values if face are
>> >> > +detected:
>> >> > + at table @option
>> >> > + at item lavfi.facedetect.nb_faces
>> >> > +Display the detected face number
>> >> > +
>> >> > + at item lavfi.facedetect.face_id.x, lavfi.facedetect.face_id.y
>> >> > +Display x and y of every faces, face_id is the face index which is
>> >> > range
>> >> > +from [0, nb_faces-1]
>> >> > +
>> >> > + at item lavfi.facedetect.face_id.w, lavfi.facedetect.face_id.h
>> >> > +Display width and height of every faces, face_id is the face index
>> >> > +which is range from [0, nb_faces-1]
>> >> > + at end table
>> >> > +
>> >> > @section oscilloscope
>> >> >
>> >> > 2D Video Oscilloscope.
>> >> > diff --git a/libavfilter/vf_libopencv.c b/libavfilter/vf_libopencv.c
>> >> > index 8128030b8c..b2d19bb241 100644
>> >> > --- a/libavfilter/vf_libopencv.c
>> >> > +++ b/libavfilter/vf_libopencv.c
>> >> > @@ -1,5 +1,6 @@
>> >> > /*
>> >> > * Copyright (c) 2010 Stefano Sabatini
>> >> > + * Copyright (c) 2020 Limin Wang
>> >> > *
>> >> > * This file is part of FFmpeg.
>> >> > *
>> >> > @@ -27,10 +28,16 @@
>> >> > #if HAVE_OPENCV2_CORE_CORE_C_H
>> >> > #include <opencv2/core/core_c.h>
>> >> > #include <opencv2/imgproc/imgproc_c.h>
>> >> > +#if HAVE_OPENCV2_OBJECTDETECT_OBJECTDETECT_C_H
>> >> > +#include <opencv2/objdetect/objdetect_c.h>
>> >> > +#else
>> >> > +#include <opencv/cv.h>
>> >> > +#endif
>> >> > #else
>> >> > #include <opencv/cv.h>
>> >> > #include <opencv/cxcore.h>
>> >> > #endif
>> >> > +
>> >> > #include "libavutil/avstring.h"
>> >> > #include "libavutil/common.h"
>> >> > #include "libavutil/file.h"
>> >> > @@ -82,6 +89,7 @@ typedef struct OCVContext {
>> >> > int (*init)(AVFilterContext *ctx, const char *args);
>> >> > void (*uninit)(AVFilterContext *ctx);
>> >> > void (*end_frame_filter)(AVFilterContext *ctx, IplImage *inimg,
>> >> > IplImage *outimg);
>> >> > + void (*postprocess)(AVFilterContext *ctx, AVFrame *out);
>> >> > void *priv;
>> >> > } OCVContext;
>> >> >
>> >> > @@ -326,18 +334,152 @@ static void
>> >> > erode_end_frame_filter(AVFilterContext
>> >> > *ctx, IplImage *inimg, IplIma
>> >> > cvErode(inimg, outimg, dilate->kernel, dilate->nb_iterations);
>> >> > }
>> >> >
>> >> > +typedef struct FaceDetectContext {
>> >> > + char *xml_model;
>> >> > + CvHaarClassifierCascade* cascade;
>> >> > + CvMemStorage* storage;
>> >> > + int nb_faces;
>> >> > + CvSeq *faces;
>> >> > + int add_roi;
>> >> > + AVRational qoffset;
>> >> > +} FaceDetectContext;
>> >> > +
>> >> > +static av_cold int facedetect_init(AVFilterContext *ctx, const char
>> >> > *args)
>> >> > +{
>> >> > + OCVContext *s = ctx->priv;
>> >> > + FaceDetectContext *facedetect = s->priv;
>> >> > + const char *buf = args;
>> >> > + double qoffset;
>> >> > +
>> >> > + if (args) {
>> >> > + facedetect->xml_model = av_get_token(&buf, "|");
>> >> > + if (!facedetect->xml_model) {
>> >> > + av_log(ctx, AV_LOG_ERROR, "failed to get %s, %s\n",
>> >> > args,
>> >> > facedetect->xml_model);
>> >> > + return AVERROR(EINVAL);
>> >> > + }
>> >> > +
>> >> > + if (buf && sscanf(buf, "|%lf", &qoffset) == 1) {
>> >> > + if (qoffset < -1.0 || qoffset > 1.0) {
>> >> > + av_log(ctx, AV_LOG_ERROR, "failed to get valid
>> >> > qoffset(%f))\n", qoffset);
>> >> > + return AVERROR(EINVAL);
>> >> > + }
>> >> > + facedetect->add_roi = 1;
>> >> > + facedetect->qoffset = av_d2q(qoffset, 255);
>> >> > + }
>> >> > + } else {
>> >> > + av_log(ctx, AV_LOG_ERROR, "failed to get
>> >> > haarcascade_frontalface_alt.xml model file\n");
>> >> > + return AVERROR(EINVAL);
>> >> > + }
>> >> > +
>> >> > + av_log(ctx, AV_LOG_VERBOSE, "xml_model: %s add_roi: %d qoffset:
>> >> > %d/%d\n",
>> >> > + facedetect->xml_model, facedetect->add_roi,
>> >> > facedetect->qoffset.num, facedetect->qoffset.den);
>> >> > +
>> >> > + facedetect->storage = cvCreateMemStorage(0);
>> >> > + if (!facedetect->storage) {
>> >> > + av_log(ctx, AV_LOG_ERROR, "cvCreateMemStorage() failed\n");
>> >> > + return AVERROR(EINVAL);
>> >> > + }
>> >> > + cvClearMemStorage(facedetect->storage);
>> >> > +
>> >> > + facedetect->cascade = (CvHaarClassifierCascade*)cvLoad(
>> >> > facedetect->xml_model, NULL, NULL, NULL );
>> >> > + if (!facedetect->cascade) {
>> >> > + av_log(ctx, AV_LOG_ERROR, "failed to load classifier
>> >> > cascade:
>> >> > %s
>> >> > \n", facedetect->xml_model);
>> >> > + return AVERROR(EINVAL);
>> >> > + }
>> >> > +
>> >> > + return 0;
>> >> > +}
>> >> > +
>> >> > +static av_cold void facedetect_uninit(AVFilterContext *ctx)
>> >> > +{
>> >> > + OCVContext *s = ctx->priv;
>> >> > + FaceDetectContext *facedetect = s->priv;
>> >> > +
>> >> > + if (facedetect->cascade)
>> >> > + cvReleaseHaarClassifierCascade(&facedetect->cascade);
>> >> > + if (facedetect->storage)
>> >> > + cvReleaseMemStorage(&facedetect->storage);
>> >> > +}
>> >> > +
>> >> > +static void set_meta_int(AVDictionary **metadata, const char *key,
>> >> > int
>> >> > idx,
>> >> > int d)
>> >> > +{
>> >> > + char value[128];
>> >> > + char key2[128];
>> >> > +
>> >> > + snprintf(value, sizeof(value), "%d", d);
>> >> > + snprintf(key2, sizeof(key2), "lavfi.facedetect.%d.%s", idx,
>> >> > key);
>> >> > + av_dict_set(metadata, key2, value, 0);
>> >> > +}
>> >> > +
>> >> > +static void facedetect_end_frame_filter(AVFilterContext *ctx,
>> >> > IplImage
>> >> > *inimg, IplImage *outimg)
>> >> > +{
>> >> > + OCVContext *s = ctx->priv;
>> >> > + FaceDetectContext *facedetect = s->priv;
>> >> > +
>> >> > + facedetect->faces = cvHaarDetectObjects(inimg,
>> >> > facedetect->cascade,
>> >> > facedetect->storage,
>> >> > + 1.25, 3, CV_HAAR_DO_CANNY_PRUNING,
>> >> > + cvSize(inimg->width/16,inimg->height/16), cvSize(0,0));
>> >> > +
>> >> > + facedetect->nb_faces = facedetect->faces ?
>> >> > facedetect->faces->total
>> >> > :
>> >> > 0;
>> >> > +}
>> >> > +
>> >> > +static void facedetect_postprocess(AVFilterContext *ctx, AVFrame
>> >> > *out)
>> >> > +{
>> >> > + OCVContext *s = ctx->priv;
>> >> > + FaceDetectContext *facedetect = s->priv;
>> >> > + AVRegionOfInterest *roi;
>> >> > + AVFrameSideData *sd;
>> >> > + AVBufferRef *roi_buf;
>> >> > + int i;
>> >> > +
>> >> > + if (facedetect->add_roi && facedetect->nb_faces > 0) {
>> >> > + sd = av_frame_new_side_data(out,
>> >> > AV_FRAME_DATA_REGIONS_OF_INTEREST,
>> >> > + facedetect->nb_faces * sizeof(AVRegionOfInterest));
>> >> > + if (!sd) {
>> >> > + return AVERROR(ENOMEM);
>> >> > + }
>> >> > + roi = (AVRegionOfInterest*)sd->data;
>> >> > + for(i = 0; i < facedetect->nb_faces; i++ ) {
>> >> > + CvRect *r = (CvRect*) cvGetSeqElem(facedetect->faces,
>> >> > i);
>> >> > +
>> >> > + roi[i] = (AVRegionOfInterest) {
>> >> > + .self_size = sizeof(*roi),
>> >> > + .top = r->y,
>> >> > + .bottom = r->y + r->height,
>> >> > + .left = r->x,
>> >> > + .right = r->x + r->width,
>> >> > + .qoffset = facedetect->qoffset,
>> >> > + };
>> >> > + }
>> >> > + }
>> >> > +
>> >> > + if (facedetect->nb_faces > 0)
>> >> > + av_dict_set_int(&out->metadata, "lavfi.facedetect.nb_faces",
>> >> > facedetect->nb_faces, 0);
>> >> > +
>> >> > + for(i = 0; i < facedetect->nb_faces; i++ ) {
>> >> > + CvRect *r = (CvRect*) cvGetSeqElem(facedetect->faces, i);
>> >> > +
>> >> > + set_meta_int(&out->metadata, "x", i, r->x);
>> >> > + set_meta_int(&out->metadata, "y", i, r->y);
>> >> > + set_meta_int(&out->metadata, "w", i, r->width);
>> >> > + set_meta_int(&out->metadata, "h", i, r->height);
>> >> > + }
>> >> > +}
>> >> > +
>> >> > typedef struct OCVFilterEntry {
>> >> > const char *name;
>> >> > size_t priv_size;
>> >> > int (*init)(AVFilterContext *ctx, const char *args);
>> >> > void (*uninit)(AVFilterContext *ctx);
>> >> > void (*end_frame_filter)(AVFilterContext *ctx, IplImage *inimg,
>> >> > IplImage *outimg);
>> >> > + void (*postprocess)(AVFilterContext *ctx, AVFrame *out);
>> >> > } OCVFilterEntry;
>> >> >
>> >> > static const OCVFilterEntry ocv_filter_entries[] = {
>> >> > - { "dilate", sizeof(DilateContext), dilate_init, dilate_uninit,
>> >> > dilate_end_frame_filter },
>> >> > - { "erode", sizeof(DilateContext), dilate_init, dilate_uninit,
>> >> > erode_end_frame_filter },
>> >> > - { "smooth", sizeof(SmoothContext), smooth_init, NULL,
>> >> > smooth_end_frame_filter },
>> >> > + { "dilate", sizeof(DilateContext), dilate_init, dilate_uninit,
>> >> > dilate_end_frame_filter, NULL },
>> >> > + { "erode", sizeof(DilateContext), dilate_init, dilate_uninit,
>> >> > erode_end_frame_filter, NULL },
>> >> > + { "smooth", sizeof(SmoothContext), smooth_init, NULL,
>> >> > smooth_end_frame_filter, NULL },
>> >> > + { "facedetect", sizeof(FaceDetectContext), facedetect_init,
>> >> > facedetect_uninit, facedetect_end_frame_filter,
>> >> > facedetect_postprocess
>> >> > },
>> >> > };
>> >> >
>> >> > static av_cold int init(AVFilterContext *ctx)
>> >> > @@ -355,6 +497,7 @@ static av_cold int init(AVFilterContext *ctx)
>> >> > s->init = entry->init;
>> >> > s->uninit = entry->uninit;
>> >> > s->end_frame_filter = entry->end_frame_filter;
>> >> > + s->postprocess = entry->postprocess;
>> >> >
>> >> > if (!(s->priv = av_mallocz(entry->priv_size)))
>> >> > return AVERROR(ENOMEM);
>> >> > @@ -383,18 +526,33 @@ static int filter_frame(AVFilterLink *inlink,
>> >> > AVFrame
>> >> > *in)
>> >> > AVFrame *out;
>> >> > IplImage inimg, outimg;
>> >> >
>> >> > + /* facedetect filter will passthrought the input frame */
>> >> > + if (strcmp(s->name, "facedetect")) {
>> >> > out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
>> >> > if (!out) {
>> >> > av_frame_free(&in);
>> >> > return AVERROR(ENOMEM);
>> >> > }
>> >> > av_frame_copy_props(out, in);
>> >> > + } else {
>> >> > + out = in;
>> >> > + }
>> >> >
>> >> > fill_iplimage_from_frame(&inimg , in , inlink->format);
>> >> > +
>> >> > + if (strcmp(s->name, "facedetect")) {
>> >> > fill_iplimage_from_frame(&outimg, out, inlink->format);
>> >> > s->end_frame_filter(ctx, &inimg, &outimg);
>> >> > fill_frame_from_iplimage(out, &outimg, inlink->format);
>> >> > + } else {
>> >> > + s->end_frame_filter(ctx, &inimg, NULL);
>> >> > + }
>> >> > +
>> >> > + if (s->postprocess) {
>> >> > + s->postprocess(ctx, out);
>> >> > + }
>> >> >
>> >> > + if (out != in)
>> >> > av_frame_free(&in);
>> >> >
>> >> > return ff_filter_frame(outlink, out);
>> >> > --
>> >> > 2.21.0
>> >> >
>> >> > _______________________________________________
>> >> > ffmpeg-devel mailing list
>> >> > ffmpeg-devel at ffmpeg.org
>> >> > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>> >> >
>> >> > To unsubscribe, visit link above, or email
>> >> > ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>> >
>> > --
>> > Thanks,
>> > Limin Wang
>> > _______________________________________________
>> > ffmpeg-devel mailing list
>> > ffmpeg-devel at ffmpeg.org
>> > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>> >
>> > To unsubscribe, visit link above, or email
>> > ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>
> --
> Thanks,
> Limin Wang
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
More information about the ffmpeg-devel
mailing list