diff --git a/cmake/external_libs/ffmpeg.cmake b/cmake/external_libs/ffmpeg.cmake
index 082e3a1fafa..d374317ead9 100644
--- a/cmake/external_libs/ffmpeg.cmake
+++ b/cmake/external_libs/ffmpeg.cmake
@@ -1,4 +1,5 @@
-set(FFMPEG_FLAGS
+if(CMAKE_SYSTEM_NAME MATCHES "Linux")
+    set(FFMPEG_FLAGS
         --disable-programs
         --disable-doc
         --disable-postproc
@@ -8,7 +9,20 @@ set(FFMPEG_FLAGS
         --disable-static
         --enable-shared
         --disable-x86asm
-        )
+        --extra-cflags="-D_FORTIFY_SOURCE=2 -fstack-protector-all"
+        --extra-ldflags="-Wl,-z,relro,-z,now")
+else()
+    set(FFMPEG_FLAGS
+        --disable-programs
+        --disable-doc
+        --disable-postproc
+        --disable-decoder=av1
+        --disable-libxcb
+        --disable-hwaccels
+        --disable-static
+        --enable-shared
+        --disable-x86asm)
+endif()
 
 set(REQ_URL "https://ffmpeg.org/releases/ffmpeg-5.1.2.tar.gz")
 set(SHA256 "87fe8defa37ce5f7449e36047171fed5e4c3f4bb73eaccea8c954ee81393581c")
@@ -36,6 +50,9 @@ else()
             LIBS avcodec avdevice avfilter avformat avutil swresample swscale
             URL ${REQ_URL}
             SHA256 ${SHA256}
+            PATCHES ${TOP_DIR}/third_party/patch/ffmpeg/CVE-2022-3964.patch
+            PATCHES ${TOP_DIR}/third_party/patch/ffmpeg/CVE-2022-3965.patch
+            PATCHES ${TOP_DIR}/third_party/patch/ffmpeg/CVE-2023-47342.patch
             CONFIGURE_COMMAND ./configure ${FFMPEG_FLAGS}
             )
 
diff --git a/third_party/patch/ffmpeg/CVE-2022-3964.patch b/third_party/patch/ffmpeg/CVE-2022-3964.patch
new file mode 100644
index 00000000000..493bf023e10
--- /dev/null
+++ b/third_party/patch/ffmpeg/CVE-2022-3964.patch
@@ -0,0 +1,72 @@
+diff -Npur ffmpeg-5.1.2/libavcodec/rpzaenc.c ffmpeg-5.1.2-change/libavcodec/rpzaenc.c
+--- ffmpeg-5.1.2/libavcodec/rpzaenc.c	2022-07-23 01:58:39.000000000 +0800
++++ ffmpeg-5.1.2-change/libavcodec/rpzaenc.c	2024-06-25 15:56:07.594394836 +0800
+@@ -205,7 +205,7 @@ static void get_max_component_diff(Block
+ 
+     // loop thru and compare pixels
+     for (y = 0; y < bi->block_height; y++) {
+-        for (x = 0; x < bi->block_width; x++){
++        for (x = 0; x < bi->block_width; x++) {
+             // TODO:  optimize
+             min_r = FFMIN(R(block_ptr[x]), min_r);
+             min_g = FFMIN(G(block_ptr[x]), min_g);
+@@ -277,7 +277,7 @@ static int leastsquares(uint16_t *block_
+         return -1;
+ 
+     for (i = 0; i < bi->block_height; i++) {
+-        for (j = 0; j < bi->block_width; j++){
++        for (j = 0; j < bi->block_width; j++) {
+             x = GET_CHAN(block_ptr[j], xchannel);
+             y = GET_CHAN(block_ptr[j], ychannel);
+             sumx += x;
+@@ -324,7 +324,7 @@ static int calc_lsq_max_fit_error(uint16
+     int max_err = 0;
+ 
+     for (i = 0; i < bi->block_height; i++) {
+-        for (j = 0; j < bi->block_width; j++){
++        for (j = 0; j < bi->block_width; j++) {
+             int x_inc, lin_y, lin_x;
+             x = GET_CHAN(block_ptr[j], xchannel);
+             y = GET_CHAN(block_ptr[j], ychannel);
+@@ -419,7 +419,9 @@ static void update_block_in_prev_frame(c
+                                        uint16_t *dest_pixels,
+                                        const BlockInfo *bi, int block_counter)
+ {
+-    for (int y = 0; y < 4; y++) {
++    const int y_size = FFMIN(4, bi->image_height - bi->row * 4);
++
++    for (int y = 0; y < y_size; y++) {
+         memcpy(dest_pixels, src_pixels, 8);
+         dest_pixels += bi->rowstride;
+         src_pixels += bi->rowstride;
+@@ -729,14 +731,15 @@ post_skip :
+ 
+             if (err > s->sixteen_color_thresh) { // DO SIXTEEN COLOR BLOCK
+                 uint16_t *row_ptr;
+-                int rgb555;
++                int y_size, rgb555;
+ 
+                 block_offset = get_block_info(&bi, block_counter);
+ 
+                 row_ptr = &src_pixels[block_offset];
++                y_size = FFMIN(4, bi.image_height - bi.row * 4);
+ 
+-                for (int y = 0; y < 4; y++) {
+-                    for (int x = 0; x < 4; x++){
++                for (int y = 0; y < y_size; y++) {
++                    for (int x = 0; x < 4; x++) {
+                         rgb555 = row_ptr[x] & ~0x8000;
+ 
+                         put_bits(&s->pb, 16, rgb555);
+@@ -744,6 +747,11 @@ post_skip :
+                     row_ptr += bi.rowstride;
+                 }
+ 
++                for (int y = y_size; y < 4; y++) {
++                    for (int x = 0; x < 4; x++)
++                        put_bits(&s->pb, 16, 0);
++                }
++
+                 block_counter++;
+             } else { // FOUR COLOR BLOCK
+                 block_counter += encode_four_color_block(min_color, max_color,
diff --git a/third_party/patch/ffmpeg/CVE-2022-3965.patch b/third_party/patch/ffmpeg/CVE-2022-3965.patch
new file mode 100644
index 00000000000..155aad1524b
--- /dev/null
+++ b/third_party/patch/ffmpeg/CVE-2022-3965.patch
@@ -0,0 +1,91 @@
+diff -Npur ffmpeg-5.1.2/libavcodec/smcenc.c ffmpeg-5.1.2-change/libavcodec/smcenc.c
+--- ffmpeg-5.1.2/libavcodec/smcenc.c	2022-07-23 01:58:39.000000000 +0800
++++ ffmpeg-5.1.2-change/libavcodec/smcenc.c	2024-06-25 17:07:00.100640653 +0800
+@@ -61,6 +61,7 @@ typedef struct SMCContext {
+         { \
+             row_ptr += stride * 4; \
+             pixel_ptr = row_ptr; \
++            cur_y += 4; \
+         } \
+     } \
+ }
+@@ -117,6 +118,7 @@ static void smc_encode_stream(SMCContext
+     const uint8_t *prev_pixels = (const uint8_t *)s->prev_frame->data[0];
+     uint8_t *distinct_values = s->distinct_values;
+     const uint8_t *pixel_ptr, *row_ptr;
++    const int height = frame->height;
+     const int width = frame->width;
+     uint8_t block_values[16];
+     int block_counter = 0;
+@@ -125,13 +127,14 @@ static void smc_encode_stream(SMCContext
+     int color_octet_index = 0;
+     int color_table_index;  /* indexes to color pair, quad, or octet tables */
+     int total_blocks;
++    int cur_y = 0;
+ 
+     memset(s->color_pairs, 0, sizeof(s->color_pairs));
+     memset(s->color_quads, 0, sizeof(s->color_quads));
+     memset(s->color_octets, 0, sizeof(s->color_octets));
+ 
+     /* Number of 4x4 blocks in frame. */
+-    total_blocks = ((frame->width + 3) / 4) * ((frame->height + 3) / 4);
++    total_blocks = ((width + 3) / 4) * ((height + 3) / 4);
+ 
+     pixel_ptr = row_ptr = src_pixels;
+ 
+@@ -145,11 +148,13 @@ static void smc_encode_stream(SMCContext
+         int cache_index;
+         int distinct = 0;
+         int blocks = 0;
++        int frame_y = cur_y;
+ 
+         while (prev_pixels && s->key_frame == 0 && block_counter + inter_skip_blocks < total_blocks) {
++            const int y_size = FFMIN(4, height - cur_y);
+             int compare = 0;
+ 
+-            for (int y = 0; y < 4; y++) {
++            for (int y = 0; y < y_size; y++) {
+                 const ptrdiff_t offset = pixel_ptr - src_pixels;
+                 const uint8_t *prev_pixel_ptr = prev_pixels + offset;
+ 
+@@ -170,8 +175,10 @@ static void smc_encode_stream(SMCContext
+ 
+         pixel_ptr = xpixel_ptr;
+         row_ptr = xrow_ptr;
++        cur_y = frame_y;
+ 
+         while (block_counter > 0 && block_counter + intra_skip_blocks < total_blocks) {
++            const int y_size = FFMIN(4, height - cur_y);
+             const ptrdiff_t offset = pixel_ptr - src_pixels;
+             const int sy = offset / stride;
+             const int sx = offset % stride;
+@@ -180,7 +187,7 @@ static void smc_encode_stream(SMCContext
+             const uint8_t *old_pixel_ptr = src_pixels + nx + ny * stride;
+             int compare = 0;
+ 
+-            for (int y = 0; y < 4; y++) {
++            for (int y = 0; y < y_size; y++) {
+                 compare |= memcmp(old_pixel_ptr + y * stride, pixel_ptr + y * stride, 4);
+                 if (compare)
+                     break;
+@@ -197,9 +204,11 @@ static void smc_encode_stream(SMCContext
+ 
+         pixel_ptr = xpixel_ptr;
+         row_ptr = xrow_ptr;
++        cur_y = frame_y;
+ 
+         while (block_counter + coded_blocks < total_blocks && coded_blocks < 256) {
+-            for (int y = 0; y < 4; y++)
++            const int y_size = FFMIN(4, height - cur_y);
++            for (int y = 0; y < y_size; y++)
+                 memcpy(block_values + y * 4, pixel_ptr + y * stride, 4);
+ 
+             qsort(block_values, 16, sizeof(block_values[0]), smc_cmp_values);
+@@ -224,6 +233,7 @@ static void smc_encode_stream(SMCContext
+ 
+         pixel_ptr = xpixel_ptr;
+         row_ptr = xrow_ptr;
++        cur_y = frame_y;
+ 
+         blocks = coded_blocks;
+         distinct = coded_distinct;
diff --git a/third_party/patch/ffmpeg/CVE-2023-47342.patch b/third_party/patch/ffmpeg/CVE-2023-47342.patch
new file mode 100644
index 00000000000..0eace366db0
--- /dev/null
+++ b/third_party/patch/ffmpeg/CVE-2023-47342.patch
@@ -0,0 +1,12 @@
+diff -Npur ffmpeg-5.1.2/libavformat/rtsp.c ffmpeg-5.1.2-change/libavformat/rtsp.c
+--- ffmpeg-5.1.2/libavformat/rtsp.c	2022-07-23 01:58:39.000000000 +0800
++++ ffmpeg-5.1.2-change/libavformat/rtsp.c	2024-06-25 16:37:03.333689422 +0800
+@@ -409,7 +409,7 @@ static void parse_fmtp(AVFormatContext *
+         if (rtsp_st->sdp_payload_type == payload_type &&
+             rtsp_st->dynamic_handler &&
+             rtsp_st->dynamic_handler->parse_sdp_a_line) {
+-            rtsp_st->dynamic_handler->parse_sdp_a_line(s, i,
++            rtsp_st->dynamic_handler->parse_sdp_a_line(s, rtsp_st->stream_index,
+                 rtsp_st->dynamic_protocol_context, line);
+         }
+     }