Il Song Han

Enhanced neuromorphic visual processing by segmented neuron for intelligent vehicle

The neuromorphic visual processing framework mimicking the biological vision system offers an alternative process into applying computer vision in everyday environment. With the growing interest for an effective approach for making detection of vulnerable road users for the purpose of safety enhancement, the proposed neuromorphic visual processing was tested on vulnerable road users such as cyclists on the road. The effectiveness of proposed neuromorphic networks of visual processing is evaluated for the vulnerable road user detection technology via maintaining the successful detection rate of over 95% without complex denoising network. The segmented neuron mixed with the rectifier enhanced the performance via extending the detection range by 33 % as well as saving the denoising process. The post enhancement with deep networks becomes flexible that further applications could be sought from incorporating neuromorphic visual processing. The early implementation demonstrated the advantages of fast and robust neuromorphic vision with either the mobile embedded systems of GPU or FPGA hardware processing, or the portable computer based emulator.

Neuromorphic visual object detection for enhanced driving safety

There have been many researches on computer vision of diverse complex vision algorithms. However, despite its effectiveness, computer vision algorithm sometimes lacks the robustness of mammalian visual system for the application in dynamic environments in vehicle driving or outdoors. We have proposed that the neuromorphic visual processing algorithm based on the biological vision system is an effective approach for making detection of human objects on the road and inside the car. The effectiveness of proposed neuromorphic networks of visual processing is evaluated for the advanced driver assistance and pedestrian safety technology via the 99% of successful detection rate. The enhanced frame based neuromorphic processing showed that further applications could be sought from incorporating neuromorphic visual processing into Driver State Monitoring for the purpose of enhancing driving safety.

Object Segmentation for Vehicle Video and Dental CBCT by Neuromorphic Convolutional Recurrent Neural Network

The neuromorphic visual processing inspired by the biological vision system of brain offers an alternative process into applying machine vision in various environments. With the emerging interests on transportation safety enhancement of Advanced Driver Assistance System or a driverless car, the neuromorphic convolutional recurrent neural networks was proposed and tested for the night-time vehicle or VRU detection. The effectiveness of proposed convolutional-recurrent neural networks of neuromorphic visual processing was evaluated successfully for the object detection without optimized complex template matching or prior denoising neural network. The real life road video dataset at night time demonstrated 98% of successful detection/segmentation rate with 0% False Positive. The robust performance of proposed convolutional-recurrent neural network was also applied successfully to the tooth segmentation of dental X-ray 3D CT including the gum region. The feature extraction was based on neuromorphic visual processing filters of either hand-cut filters mimicking the visual cortex experimentation or the auto-encoder filter trained by partial X-ray images. The consistent performance of either hand-cut filters or the small auto-encoder filters demonstrated the feasibility of real-time and robust neuromorphic vision implemented by either the small embedded system or the portable computer.

Bio-inspired computing of vision — Fuzzy and neuromorphic processing

This paper describes the early vision of bio-inspired neuromorphic system enhanced by fuzzy processing, mimicking the primitive behaviour of visual cortex. The proposed bio-inspired system exhibits the biologically plausible function of mimicking the cats visual cortex experimentation of Hubel and Wiesel. The neuromorphic implementation of vision is inspired by the directional visual signal selectivity of cortex and the CMOS spiking neuron based on Hodgkin-Huxley formalism. The membership function is introduced to improve the early vision and the improvement is demonstrated for robust applications, based on the recognition of human head figures.