Vehicle Detection and Classification using Improved Faster Region Based Convolution Neural Network

Abstract

Over the last few years, object detection has arisen as a powerful tool for developing reliable systems like detection of vehicles using computer vision. Convolutional neural network can be used to employ feature synthesis methods to concatenate low level and high level features as well as to detect vehicles of different size and scale. In this paper, an improved faster R-CNN method has been proposed. The proposed method has been evaluated using FLIR_ADAS dataset for both thermal and RGB images. Experiments have been performed on thermal images and results demonstrated significant increase in accuracy as well as non-ambiguous detections as compared to conventional Faster R-CNN method.