High Dynamic Range Video

Natural scenes usually have a range of brightness values that exceeds the capabilities of digital capturing devices by far. Traditional “low dynamic range” (LDR) imaging devices often cannot capture their full dynamic range, especially that of back-lit scenes. This leads to under- and overexposed pixels in the captured images, and information on brightness differences between these pixels is lost.

The most popular approach to overcome this limitation is using a set of LDR images captured in quick sequence at different exposure settings. Each LDR image then captures one facet of the scene's dynamic range. While low exposure images show more contrast in bright areas, the images captured with high exposure reveal all details in darker regions. When fused together, an HDR image is created that covers the full dynamic range of the scene.

A dark, medium and bright exposure of a scene with direct sunlight. The picture on the bottom right is the tone mapped HDR image created by merging multiple exposures.

This approach can be extended to creating high dynamic range videos. In the simplest case, an HDR video can be produced by generating HDR images in quick sequence. However, a great portion of the required capturing and processing time can be saved by exploiting the temporal coherence in a video. The result is HDR video captured in real-time.

The links below provide examples of two scenes captured in LDR as well as HDR. The average frame rate is approximately 4 frames per second, but our current work focuses on improving the algorithms and their implementation to achieve near TV frame rates (~25 fps).


                                       Low Dynamic Range                                                                        Low Dynamic Range
                                       High Dynamic Range                                                                      High Dynamic Range



Flicker Reduction in HDR Video

The following link provides various downloadable versions of five HDR videos. They have been tone mapped with three different tone mapping operators each. For each combination, an original version and and one where image flicker was removed is available.