Suppressing Rain Streak Weather Noise for Recognizing Overturned Accidental Car Vehicle License Plate

Vehicle License Plate Recognition (VLPR) system is widely used in real time gate mechanism, parking lots, toll plaza etc. Such system fails when there is an occurrence of rain weather noise over car License Plate. This paper proposes new algorithm for efficient weather noise removal incorporated along with YOLO for efficient character recognition. The RS Removal for VLPR system is named as RSRVLPR. The existing spatial RS removal scheme uses guided filter and Deep Learning (DL). However, the scheme is inefficient due to time complexity. The proposed RSRVLPR scheme uses Prelu based Detail Layer (PDL) based on Deep Learning that performs the removal of foreground rain streak by super imposing the Detail Layer over VLP. The learning process is efficient as the Prelu based activation function models the non-linearity. After the removal of rain streak, the License Plate (LP) of any orientation is detected and recognized using You Only Look Once (YOLO). The proposed RSRVLPR is implemented in Anaconda using Jupyter Notebook with co-lab. The experiment results show that the proposed PDL scheme distinguishes the rain streaks from background to get the clear de-rain image as an output to YOLO. The performance of the PDL based RSRVLPR is evaluated with the state-of-art methods in terms of recognition accuracy for both synthetic and raw data and found to give accuracy of 84% in RSR, 91% for LP detection and 93% for LP recognition.