Reflection Removal for In-Vehicle Black Box Videos

Christian Simon, In Kyu Park; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 4231-4239

In-vehicle black box camera becomes an popular equipment in many countries for security monitoring and event capturing. The readability of video contents is the most important capability, which is, however, often degraded due to the reflection on the windscreen. In this paper, we propose a novel method to remove the reflection on the windscreen in the in-vehicle black box videos. The main idea is to exploit spatio-temporal coherence of the reflection which shows that a vehicle moves forward while the reflection layer of internal object remains static. The average image prior is introduced by imposing a heavy-tail distribution with higher peak. The two layered scene is the base of the separation model. In order to remove the reflection, a cost non-convex function is developed based on this property and optimized. Experimental results demonstrate that the proposed approach successfully separates the layers in real black box videos.