Vehicle Detection on Sanctuaries Using Spatially Distributed Convolutional Neural Network

Abstract

Nowadays, detection of vehicles from images captured using web camera is becoming an important focus in the research field of image processing. A common issue inside a wildlife sanctuary is that the possibility of vehicle moving in wrong way or getting into some problem is high. This work proposes an algorithm for detection of vehicles using CNN (Convolutional neural network) on the basis of SDP (spatially distributed pooling). Finite length feature vector is developed by sampling various sized behavioural pattern using SDP. Thus effect of detection can be improved by avoiding distortion of different sized images. Also, Normed slope (NS) method is proposed with more number of threshold as an algorithm for image pre-processing. Using NS, retains the object edge which might be disturbed in the infrastructure. Computational cost for extraction of candidate objects is lesser as only a limited candidate windows are generated. Results of experiments reveal that the SDP based CNN can work well on multiple sized input images thereby improving the effect of detection.