Implementation of a high-performance portable real-time dehazing system using the DCP algorithm

This document presents a portable, high performance system for real-time image dehazing using the DCP algorithm. The software runs on a CPU-GPU heterogeneous environment in a development platform from NVIDIA with support for CUDA. Three different window sizes for the DCP algorithm were explored. These low values allow a high throughput and reduce the halo artifacts but results in low-luminosity frames which negatively impacts quality. To solve this, a gamma correction was introduced as a final stage. The coefficient for this filter can vary and is a value which can be manually modified with feedback from the users, depending on their preferences. The picture resolution utilized was
$1280\times 720$
pixels, which the system was able to process at a frame rate of 40 fps using a window size of 3. The image capture can come either from an USB camera or from a video file. The output can be store as a file or it can be wirelessly transmitted to a mobile device using the RTP protocol.