Life-like Image Processing for Small Target Motion Detection in Cluttered Dynamic Environments

Wang, Hongxin (2019) Life-like Image Processing for Small Target Motion Detection in Cluttered Dynamic Environments. PhD thesis, University of Lincoln.

Life-like Image Processing for Small Target Motion Detection in Cluttered Dynamic Environments
PhD Thesis hONGXIN wANG.pdf - Whole Document

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


Discriminating targets moving against a cluttered background is a huge challenge for future robotic vision systems, let alone detecting a target as small as one or a few pixels. As a source of inspiration, insects are quite apt at searching for mates and tracking prey – which always appear as small dim speckles in the visual field. The exquisite sensitivity of insects for small target motion, as revealed recently, is coming from a class of specific neurons called small target motion detectors (STMDs). Some of the STMDs have also demonstrated direction selectivity which means these STMDs respond strongly only to their preferred motion direction. Build a quantitative STMD model is the first step for not only further understanding of the biological visual system, but also providing robust and economic solutions of small target detection for an artificial visual system. This research aims to explore STMD-based image processing methods for small target motion detection against cluttered dynamic backgrounds. The major contributions are summarized as follows.

Three STMD-based neural models are proposed in this research named as directionally selective STMD(DSTMD), STMD Plus and Feedback STMD, respectively. The DSTMD systematically models and studies direction selectivity of the STMD neurons, meanwhile provides with unified and rigorous mathematical description. Specifically, in the DSTMD, a new correlation mechanism is introduced for direction selectivity via correlating signals relayed from two pixels. Then, a lateral inhibition mechanism is implemented on the spatial field for size selectivity of the STMD neurons. Finally, a population vector algorithm is used to encode motion direction of small targets. Extensive experiments showed that the proposed DSTMD not only is in accord with current biological findings, i.e. showing directional preferences, but also works reliably in detecting small targets against cluttered backgrounds.

The STMD Plus is developed to discriminate small targets from small-target-like background features (named as fake features) by integrating motion information with directional contrast. More precisely, the STMD Plus is composed of four subsystems – ommatidia, motion pathway, contrast pathway and mushroom body. Compared to existing STMD-based models, the additional contrast pathway extracts directional contrast from luminance signals to eliminate false positive background motion. The directional contrast and the extracted motion information by the motion pathway are integrated in the mushroom body for small target discrimination. The experimental results demonstrated the significant and consistent improvements of the proposed visual system model over existing STMD-based models against fake features.

The Feedback STMD is also designed to filter out fake features by introducing a new feedback mechanism. Specifically, the model output is first temporally delayed then applied to the previous neural layer to construct a feedback loop. By subtracting the feedback signal from the inputs of the STMDs, the background fake features are largely suppressed. Experimental results show that the developed feedback neural model achieves better performance than the existing STMD-based models in discriminating small targets from complex backgrounds.

Divisions:College of Science > School of Computer Science
ID Code:47485
Deposited On:06 Dec 2021 10:59

Repository Staff Only: item control page