Accident Scene Image Identification Technique using Convolutional Neural Network

Abstract

ABSTRACT Building intelligent software that can effectively detect accident scene with the help of Google map has suffered set back because of the poor ability of the currently used software to effectively detect, identify and classify accident scene images from non accident scene images. Hence there is need for a better technique of implementing this software. In this paper, Convolutional neural networks (CNN) which is a part of deep learning algorithm was used to provide a better classification technique that any software to be developed for the purpose of detecting accident scene image can adopt. The algorithm was tested on 4000 accident scene images with other kind of images (cats and dogs) by adopting models of other researchers. In this paper, classification accuracy and Mean Squared Error (MSE) were used to evaluate the algorithm in identifying accident scene images accurately. The result was further presented using a graph of MSE against a number of trained epochs. The result of the experiment shows accuracy in the image classification and identification