Hardware Implementation of Combining Image Enhancement and ROI Extraction for Night Time Images

R. Nivedha PG Student / ECE Kings College of Engineering, Punalkulam Thanjavur, Tamil Nadu Email:[email protected] W. Newton David Raj M.Tech., Assistant Professor / ECE Kings College of Engineering, Punalkulam Thanjavur, Tamil Nadu Email: [email protected] -------------------------------------------------------------------ABSTRACT--------------------------------------------------------------In nighttime images, vehicle detection is a tricky task because of low contrast and luminosity. To enhance those images by normal MSR gives the apparent data loss because it uses a logarithmic function. To minimize the data loss, the proposed Multi-Scale Retinex Color Restoration (MSRCR) is used. It uses customized sigmoid function which result in preserve areas with normal lighting and suppress noise speckles in very low light areas and also it increases the Peak Signal to Noise Ratio (PSNR) of the image. In this paper, we combine a nighttime image enhancement approach based on Multi-Scale Retinex Color Restoration (MSRCR) and a novel region of interest (ROI) extraction approach that fuses object proposals and vehicle light detection together to enhance images and to extract accurate ROIs for accurate nighttime vehicle detection. Since Field Programmable Gate Array (FPGA) performs the high speed processing required for more accurate detection, FPGA technology will be used for the implementation of Artificial Neural Network (ANN) classifier. Our proposed method can detect blurred and partly occluded vehicles, as well as vehicles in a variety of sizes, locations and backgrounds. --------------------------------------------------------------------------------------------------------------------------------------------------