Use of Optimized Convolutional Neural Network (OCNN) for Object Detection

Abstract

In the daily life routine, Visually Impaired (VI) persons are mostly suffering from the trouble of physical movements. As getting the exact information regarding the objects ahead and accordingly move further without colliding or not getting harmed is a greatest challenge for them in the indoor environments also. The purpose of this research is to propose model for object detection in an indoor environment for visually impaired person. Initially, the input image from real-world scenario is given to pre-processing via wiener filtering. Subsequently, segmentation is carried out by the novel proposed Multi-Kernel K-means clustering model and SURF (Speeded-Up Robust Features), SIFT (Scale-Invariant Feature Transform), Shape based features via canny edges & Gradient features via HOG (Histogram of Oriented Gradients) were extracted from segmented image. The optimal features are selected from extracted features by a new hybrid algorithm referred as Particle Hybridized SeaLion Optimization Algorithm (PHSLA) and selected optimal features are classified using Optimized Convolutional Neural Network (OCNN) for detecting the object. It was observed from the comparative study & performance evaluation that OCNN and PHSLA received 98% accuracy in identification of object and helping system to become more stable and reliable