Bio-inspired Collision Detection with Motion Cues Enhancement in Dim Light Environments

Lei, Fang (2022) Bio-inspired Collision Detection with Motion Cues Enhancement in Dim Light Environments. PhD thesis, University of Lincoln.

Bio-inspired Collision Detection with Motion Cues Enhancement in Dim Light Environments
PhD Thesis
Bio-inspired Collision Detection with Motion Cues Enhancement in Dim Light Environments
Electronic submission form
Lei, Fang – Computer Science – May 2022.pdf - Whole Document

[img] Microsoft Word
Research Electronic Thesis Submission FormFang Lei.docx - Supplemental Material

Item Type:Thesis (PhD)
Item Status:Live Archive


Detecting looming objects robustly and timely is a huge challenge for artificial vision systems in complex natural scenes, including dim light scenes. Insects have evolved remarkable capacities in collision detection despite their tiny eyes and brains. The locusts’ LGMD1 neuron shows strong looming-sensitive property for both light and dark objects, which is a source of inspiration for developing collision detection systems. Furthermore, specialized visual processing strategies in nocturnal animals’ brains can provide inspiration for detecting faint motion like dim-light collision detection when challenged with low light conditions. This research aims to explore theLGMD1 based collision detection methods, adaptive low-light image enhancement methods, biologically-inspired solutions for enhancing faint motion cues as well as collision detection methods in low light conditions. The major contributions are summarized as follows.

A new visual neural system model (LGMD1) is developed, which applies a neural competition mechanism within a framework of separated ON and OFF pathways to shut off the translating response. The competition-based approach responds vigorously to monotonous ON/OFF responses resulting from a looming object. However, it does not respond to paired ON-OFF responses that result from a translating object, thereby enhancing collision selectivity. Moreover, a complementary denoising mechanism ensures reliable collision detection. To verify the effectiveness of the model, we have conducted systematic comparative experiments on synthetic and real datasets. The results show that our method exhibits more accurate discrimination between looming and translational events—the looming motion can be correctly detected. It also demonstrates that the proposed model is more robust than comparative models.

A framework is proposed for adaptively enhancing low-light images, which implements the processing of dark adaptation with proper adaptation parameters in R,G and B channels separately. Specifically, the dark adaptation processing consists of a series of canonical neural computations, including the power law adaptation, divisive normalization and adaptive rescaling operations. Experimental results show that the proposed bioinspired dark adaptation framework is more efficient and can better preserve the naturalness of the image compared with several representative low light image enhancement methods.

A dim-light motion cues enhancement (DLMCE) model is designed for extracting extremely faint motion cues. This model integrates dark-adaptation, spatio-temporal constraint and neural summation mechanisms, which are achieved with canonical neural computations and neural summation in temporal and spatial domains, to enhance faint motion cues. With the DLMCE model, the image intensity and contrast are first increased by the dark adaptation processing, then the strong motion cues are extracted by the spatio-temporal constraint strategy, and these motion cues are further enhanced by neural summation mechanisms. Experimental results have demonstrated that the presented DLMCE model outperforms the existing methods for dim-light motion cues enhancement, and faint motion cues can be successfully detected in consecutive frames efficiently. As demonstrated in the experiments, the proposed DLMCE model provides a robust and effective solution for autonomous systems in detecting moving objects under low light conditions.

A bio-inspired collision detection model is developed for detecting looming objects in dim light environments. The model combines the DLMCE model with the classical four-layered LGMD1 model to detect dimly illuminated approaching objects. To verify the effectiveness of the model, we have conducted comparative experiments on real looming datasets. The results have demonstrated that the proposed bio-inspired collision detection model can correctly recognize looming objects under low light conditions since the DLMCE model enhances the faint looming cues.

Keywords:Collision detection, motion cues, dim light environment, dim-light motion cues, visual neural system model, looming objects
Subjects:G Mathematical and Computer Sciences > G400 Computer Science
Divisions:College of Science > School of Computer Science
ID Code:50537
Deposited On:25 Aug 2022 09:12

Repository Staff Only: item control page