TSOM: Small object motion detection neural network inspired by avian visual circuit.

Area of Science:

• Computational Neuroscience
• Machine Vision
• Avian Visual Systems

Background:

• Detecting small moving objects from overhead views in complex backgrounds is a significant challenge for machine vision.
• The avian visual system, particularly the Retina-OT-Rt circuit, excels at processing motion in complex aerial environments.

Purpose of the Study:

• To develop a biologically inspired algorithm for small object motion detection.
• To propose a novel neural network model based on the avian Retina-OT-Rt visual circuit.

Main Methods:

• Mathematical modeling of the avian Retina-OT-Rt visual circuit's biological mechanisms.
• Development of the Tectum Small Object Motion (TSOM) detection neural network with distinct functional layers (retina, SGC dendritic, SGC Soma, Rt).
• Validation through pigeon neurophysiological experiments and analysis of image sequence data.

Main Results:

• The proposed TSOM neural network effectively mimics topographic projection, spatial-temporal encoding, motion feature selection, and multi-directional motion integration.
• Experimental results demonstrate the biological interpretability and efficacy of TSOM in extracting motion features from complex high-altitude backgrounds.

Conclusions:

• The TSOM neural network provides a biologically plausible and effective solution for small object motion detection in challenging visual conditions.
• This research bridges avian visual neuroscience and machine vision, offering new avenues for advanced motion detection algorithms.