Preprocessed Faster RCNN for Vehicle Detection

Abstract

This paper presents a pre-processed faster region convolution neural network (faster RCNN) for the purpose of on-road vehicle detection. The system introduces a preprocessing pipeline on faster RCNN. The preprocessing method is for the improvement on training and detection speed of Faster RCNN. A preprocessing lane detection pipeline based on the Sobel edge operator and Hough Transform is used to detect lanes. A Rectangular region is then extracted from lane coordinates which is a reduced region of interest (ROI). Results show that the proposed method improves the training speed of faster RCNN when compared to faster RCNN without preprocessing.