Bio-inspired neuromorphic visual processing with neural networks for cyclist detection in vehicle's blind spot and segmentation in medical CT images

Abstract

Neuromorphic visual processing inspired by the biological vision system of brain offers an alternative process into applying machine vision in everyday environment. With the growing demand for an effective detection method of moving objects on the road for the purpose of transportation safety enhancement, the proposed neuromorphic visual processing was tested on the vehicle's blind spot cyclist detection. The effectiveness of proposed convolutional-recurrent mixed networks of neuromorphic visual processing is evaluated for the cyclist detection without optimized complex template matching or denoising layers of neural network. The new feature extraction by integrating both hand-cut convolution filters and autoencoder is designed for processing the noisy image including the 3-dimensional tooth segmentation in the gum region. The proposed mixed feature extraction by the hand-cut filter and auto-encoder demonstrates the cyclist detection rate of 98% for vehicles on the road or the successful segmentation for medical CT images of dental X-ray 3D CT including the gum region or brain CT in BRATS data sets.