diff --git a/modules/calib3d/src/stereosgbm.cpp b/modules/calib3d/src/stereosgbm.cpp
index 74d5068ac5..ff033e7a0c 100644
--- a/modules/calib3d/src/stereosgbm.cpp
+++ b/modules/calib3d/src/stereosgbm.cpp
@@ -2074,41 +2074,44 @@ void SGBM3WayMainLoop::impl(const Range& range) const
             {
                 d = 0;
 #if (CV_SIMD || CV_SIMD_SCALABLE)
-                horPassCostVolume+=x;
-                int thresh = (100*min_cost)/(100-uniquenessRatio);
-                v_int16 thresh_reg = vx_setall_s16((short)(thresh+1));
-                v_int16 d1 = vx_setall_s16((short)(best_d-1));
-                v_int16 d2 = vx_setall_s16((short)(best_d+1));
-                v_int16 eight_reg = vx_setall_s16((short)VTraits<v_int16>::vlanes());
-                v_int16 cur_d = vx_load(idx_row);
-                v_int16 mask;
+                if(uniquenessRatio<100)
+                {
+                    horPassCostVolume+=x;
+                    int thresh = (100*min_cost)/(100-uniquenessRatio);
+                    v_int16 thresh_reg = vx_setall_s16(saturate_cast<short>(thresh+1));
+                    v_int16 d1 = vx_setall_s16((short)(best_d-1));
+                    v_int16 d2 = vx_setall_s16((short)(best_d+1));
+                    v_int16 eight_reg = vx_setall_s16((short)VTraits<v_int16>::vlanes());
+                    v_int16 cur_d = vx_load(idx_row);
+                    v_int16 mask;
 
-                for( ; d <= D - 2*VTraits<v_int16>::vlanes(); d+=2*VTraits<v_int16>::vlanes() )
-                {
-                    mask = v_and(v_lt(vx_load_aligned(horPassCostVolume + d), thresh_reg), v_or(v_lt(cur_d, d1), v_gt(cur_d, d2)));
-                    cur_d = v_add(cur_d, eight_reg);
-                    if( v_check_any(mask) )
-                        break;
-                    mask = v_and(v_lt(vx_load_aligned(horPassCostVolume + d + VTraits<v_int16>::vlanes()), thresh_reg), v_or(v_lt(cur_d, d1), v_gt(cur_d, d2)));
-                    cur_d = v_add(cur_d, eight_reg);
-                    if( v_check_any(mask) )
-                        break;
-                }
-                if( d <= D - 2*VTraits<v_int16>::vlanes() )
-                {
-                    horPassCostVolume-=x;
-                    continue;
-                }
-                if( d <= D - VTraits<v_int16>::vlanes() )
-                {
-                    if( v_check_any(v_and(v_lt(vx_load_aligned(horPassCostVolume + d), thresh_reg), v_or(v_lt(cur_d, d1), v_gt(cur_d, d2)))) )
+                    for( ; d <= D - 2*VTraits<v_int16>::vlanes(); d+=2*VTraits<v_int16>::vlanes() )
+                    {
+                        mask = v_and(v_lt(vx_load_aligned(horPassCostVolume + d), thresh_reg), v_or(v_lt(cur_d, d1), v_gt(cur_d, d2)));
+                        cur_d = v_add(cur_d, eight_reg);
+                        if( v_check_any(mask) )
+                            break;
+                        mask = v_and(v_lt(vx_load_aligned(horPassCostVolume + d + VTraits<v_int16>::vlanes()), thresh_reg), v_or(v_lt(cur_d, d1), v_gt(cur_d, d2)));
+                        cur_d = v_add(cur_d, eight_reg);
+                        if( v_check_any(mask) )
+                            break;
+                    }
+                    if( d <= D - 2*VTraits<v_int16>::vlanes() )
                     {
                         horPassCostVolume-=x;
                         continue;
                     }
-                    d+=VTraits<v_int16>::vlanes();
+                    if( d <= D - VTraits<v_int16>::vlanes() )
+                    {
+                        if( v_check_any(v_and(v_lt(vx_load_aligned(horPassCostVolume + d), thresh_reg), v_or(v_lt(cur_d, d1), v_gt(cur_d, d2)))) )
+                        {
+                            horPassCostVolume-=x;
+                            continue;
+                        }
+                        d+=VTraits<v_int16>::vlanes();
+                    }
+                    horPassCostVolume-=x;
                 }
-                horPassCostVolume-=x;
 #endif
                 for( ; d < D; d++ )
                 {
diff --git a/modules/calib3d/test/test_stereomatching.cpp b/modules/calib3d/test/test_stereomatching.cpp
index 9ea16110b1..910dd50a12 100644
--- a/modules/calib3d/test/test_stereomatching.cpp
+++ b/modules/calib3d/test/test_stereomatching.cpp
@@ -952,6 +952,18 @@ TEST(Calib3d_StereoSGBM, deterministic) {
 
 }
 
+TEST(Calib3d_StereoSGBM, regression_29761)
+{
+    Mat leftImg(256, 256, CV_8UC1), rightImg(256, 256, CV_8UC1);
+    randu(leftImg, Scalar(0), Scalar(255));
+    randu(rightImg, Scalar(0), Scalar(255));
+
+    Ptr<StereoSGBM> sgbm = StereoSGBM::create( 0, 64, 5, 100, 200, 1, 31, 100, 0, 0, StereoSGBM::MODE_SGBM_3WAY);
+    Mat leftDisp;
+    sgbm->compute( leftImg, rightImg, leftDisp);
+    EXPECT_EQ(countNonZero(leftDisp != -StereoMatcher::DISP_SCALE), 0);
+}
+
 TEST(Calib3d_StereoSGBM_HH4, regression)
 {
     String path = cvtest::TS::ptr()->get_data_path() + "cv/stereomatching/datasets/teddy/";