CNN Real-Time Detection of Vandalism Using a Hybrid -LSTM Deep Learning Neural Networks

Abstract

Vandalism is a deliberate damage to property by humans and it has become rampant in the engineering fields. The activity results into huge financial and social loses and the vice is declared when human image is detected in the restricted area without authority to cause an unauthorized change in a predetermined scene that could be vandalized. This act requires an automated real-time detection of the presence of the vandal so that he can be stopped from damaging the property. Human Image recognition process is the best method for detection of vandalism. In this research paper, we propose a deep learning architecture combining Convolutional Neural Networks and Long Short Term memory (CNN-LSTM) which has the ability to exhaust spatial relationship and temporal prediction of the output. The main objective of this research work is to develop, train, test and validate CNN-LSTM against CNN and LSTM models to prove the superiority of the proposed model in image recognition. Image detection is achieved by feeding the images captured by installed image sensors (CCD camera) to a hybrid neural network classifier which is trained to recognize human images. The CNN-LSTM hybrid approach not only improves the predictive accuracy of image recognition from raw data but also reduces the computational complexity. The model is trained and tested with image-Net dataset which is the largest clean image dataset for vision research. Results show that the proposed model is able to achieve a training accuracy of 98% while a standalone CNN achieved 88%. The result show that the hybrid model is superior.