Hardware Architecture for Stitching Images with Haze Removal Function

Clear and wide-angle images are essential to detect and trace objects for applications such as CCTV and unmanned vehicles. Fog, haze, and rain causes problem in getting clear image and lowers the detection rate of objects. To remove haze, we use median dark channel prior which can reduce the complexity of hardware. Image stitching stitches a number of images from multiple cameras into single wide-angle image like a panorama view. It is more efficient for surveillance to use single wide-angle image than multiple images. Stitched image is generated by several steps, feature point extraction, matching, and brightness correction. For real-time application, software implementation is not feasible due to its large complexity and long computation. Therefore, we exploit parallel processing architecture of hardware to implement haze removal and stitching. FPGA implementation shows that the performance is 30fps at 1280x720 resolution which is suitable for real-time applications.