Modelling Drosophila motion vision pathways for decoding the direction of translating objects against cluttered...

Area of Science:

• Computational Neuroscience
• Animal Vision
• Bio-inspired Computing

Background:

• Detecting translating objects against cluttered, moving backgrounds is a significant challenge in computer vision.
• Fruit flies (Drosophila) offer a biological model for understanding motion perception due to their efficient visual systems in complex environments.
• Existing models often struggle with background noise and accurately discerning target motion direction.

Purpose of the Study:

• To investigate the motion vision pathways of Drosophila melanogaster.
• To develop a bio-plausible computational model of insect motion perception.
• To achieve robust decoding of object direction in cluttered, dynamic visual scenes.

Main Methods:

• Developed a computational model incorporating bio-plausible ON and OFF visual pathways.
• Modeled wide-field horizontal-sensitive (HS) and vertical-sensitive (VS) systems.
• Integrated spatiotemporal dynamics, motion pre-filtering, and local correlator ensembles within ON/OFF pathways to suppress background noise.

Main Results:

• The model successfully generates direction-selective and direction-opponent responses in a feed-forward manner.
• Demonstrated robust direction selectivity for translating objects against cluttered backgrounds.
• The model effectively suppresses irrelevant background motion and distractors, enhancing dynamic response.

Conclusions:

• The proposed neural system model accurately decodes object direction using global responses from HS and VS systems.
• The model's effectiveness was experimentally verified, showing preference for faster, higher-contrast, and larger targets.
• This research provides insights into insect motion perception strategies applicable to bio-inspired artificial vision systems.