Merge pull request #28692 from YangGuanyuhan:fix-lightglue-assertion-fail

dnn: fix dst_dp assertion in broadcast for size-1 dims causing crash in lightglue.onnx model #28692

The original assertion CV_Assert(dst_dp == 1) does not handle valid cases where the innermost dimension size is 1 like [10, 5, 1], resulting in dst_dp == 0.

This occurs during broadcasting in LightGlue ONNX model and leads to assertion failure.

Allow dst_dp == 0 for size-1 dimensions to handle this edge case correctly.

### Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
      Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake

modules/core/src/matrix_transform.cpp

@@ -890,7 +890,7 @@ void broadcast(InputArray _src, InputArray _shape, OutputArray _dst) {
     if (_flatten_for_broadcast(2, max_ndims, all_ndims, orig_shapes, flatten_shapes, flatten_steps)) {
         size_t src_dp = flatten_steps[0][max_ndims - 1];
         size_t dst_dp = flatten_steps[1][max_ndims - 1];
-        CV_Assert(dst_dp == 1);
+        CV_Assert(dst_dp == 1 || dst_dp == 0);
         CV_Assert(max_ndims >= 2); // >= 3?
         size_t rowstep_src = flatten_steps[0][max_ndims - 2];
         size_t rowstep_dst = flatten_steps[1][max_ndims - 2];

modules/core/test/test_arithm.cpp

@@ -2818,6 +2818,39 @@ TEST(BroadcastTo, basic) {
     }
 }
 
+TEST(BroadcastTo, regression_dst_dp_zero_when_last_dim_is_one)
+{
+    std::vector<int> shape_src{10, 1, 1};
+    std::vector<float> data_src(10);
+    for (int i = 0; i < 10; ++i)
+    {
+        data_src[i] = static_cast<float>(i + 1);
+    }
+    Mat src(static_cast<int>(shape_src.size()), shape_src.data(), CV_32FC1, data_src.data());
+
+    std::vector<int> shape_dst{10, 5, 1};
+    Mat dst;
+
+    // Regression for broadcast() path where the innermost destination dimension is 1
+    // and flattened destination step can legitimately be 0.
+    ASSERT_NO_THROW(broadcast(src, shape_dst, dst));
+
+    EXPECT_EQ(dst.dims, 3);
+    EXPECT_EQ(dst.size[0], 10);
+    EXPECT_EQ(dst.size[1], 5);
+    EXPECT_EQ(dst.size[2], 1);
+    EXPECT_EQ(dst.type(), CV_32FC1);
+
+    for (int i = 0; i < shape_dst[0]; ++i)
+    {
+        for (int j = 0; j < shape_dst[1]; ++j)
+        {
+            int idx[] = {i, j, 0};
+            EXPECT_FLOAT_EQ(dst.at<float>(idx), static_cast<float>(i + 1));
+        }
+    }
+}
+
 TEST(Core_minMaxIdx, regression_9207_2)
 {
     const int rows = 13;