Congratulations to Sebastian for completing his Master's Thesis!

Optically Enhancing Event-Based Vision

Abstract

As an asynchronous imaging sensor with integrated change-detection capabilities, EBS can detect temporal changes in scene brightness as they are happening. However, its capability to detect these changes is implemented in electronics and limits the ability to detect low-contrast objects and generates significant noise, both of which hinder the ability to detect objects of interest. Optical spatial high-pass filtering, and optical photocurrent biasing is introduced to address two key Event-Based Sensor (EBS) limitations in low contrast sensitivity and low-light shot noise. The low-contrast sensitivity of EBS may be improved with optical, coherent high-pass spatial frequency filtering (HPF). This provides contrast amplification to imaged features associated with sharp edges and fine details, increasing the probability of detection of low-contrast moving objects. We present optical HPF to improve detection performance in Event-Based systems. This approach demonstrates a hardware-based solution to improving overall Event-Based system contrast sensitivity by pre-filtering imaged scenes in the optical domain. Measurements show that objects containing features with contrasts as low as 3.53% are discernable, which enables object detection with triple the sensitivity of the standalone EBS. Regarding the noise problem, the sensor’s usability in low-light conditions is greatly limited by shot noise from photons and from sensor transistor circuitry. Particularly, low illumination levels signals are limited by parasitic dark current which becomes confused for objects of interest and limits overall signal-to-noise ratio (SNR). EBS noise behavior as a function of illumination level is well documented in literature and suggests that there exists an optimal background illumination level for noise minimization. This work introduces the development of passive noise correction via a synthetic injection of a spatially homogeneous light field into the EBS’ imaging path. This circumvents significant low-level illumination noise with up to 5x boost in normalized signal minus noise ratio, prior to post-processed noise filtering. This work may guide future development of optical improvements to the EBS and suggests improved practices for enhancing overall system performance.

Committee:

• Dr. Amit Ashok (Chair)
• Dr. Ron Driggers
• Dr. Jon Koshel

Previous post
Congratulations to Joe for defending his dissertation! Next post
Congratulations to Trevor for completing his Master's Thesis!