Search

Differential Evolution (DE) is a simple, yet highly competitive real parameter optimizer in the family of evolutionary algorithms. A significant contribution of its robust performance is attributed to its control parameters, and mutation strategy employed, proper settings of which, generally lead to good solutions. Finding the best parameters for a given problem through the trial and error method is time consuming, and sometimes impractical. This calls for the development of adaptive parameter control mechanisms. In this work, we investigate the impact and efficacy of adapting mutation strategies with or without adapting the control parameters, and report the plausibility of this scheme. Backed with empirical evidence from this and previous works, we first build a case for strategy adaptation in the presence as well as in the absence of parameter adaptation. Afterwards, we propose a new mutation strategy, and an adaptive variant SA-SHADE which is based on a recently proposed self-adaptive memory based variant of Differential evolution, SHADE. We report the performance of SA-SHADE on 28 benchmark functions of varying complexity, and compare it with the classic DE algorithm (DE/Rand/1/bin), and other state-of-the-art adaptive DE variants including CoDE, EPSDE, JADE, and SHADE itself. Our results show that adaptation of mutation strategy improves the performance of DE in both presence, and absence of control parameter adaptation, and should thus be employed frequently.

Video analytics is emerging as a high potential area supplementing intelligent transportation systems (ITSs) with wide ranging applications from traffic flow analysis to surveillance. Object detection and classification, as a sub part of a video analytical system, could potentially help transportation agencies to analyze and respond to traffic incidents in real time, plan for possible future cascading events, or use the classification data to design better roads. This work presents a specialized vehicle classification system for urban environments. The system is targeted at the analysis of vehicles, especially trucks, in urban two lane traffic, to empower local transportation agencies to decide on the road width and thickness. The main thrust is on the accurate classification of the vehicles detected using an evolutionary algorithm. The detector is backed by a differential evolution (DE) based discrete parameter optimizer. The authors show that, though employing DE proves expensive in terms of computational cycles, it measurably improves the accuracy of the classification system.

Swarm intelligence algorithms are inherently parallel since different individuals in the swarm perform independent computations at different positions simultaneously. Hence, these algorithms lend themselves well to parallel implementations thereby speeding up the optimization process. FireWorks Algorithm (FWA) is a recently proposed swarm intelligence algorithm for optimization. This work investigates the scalability of the parallelization of the Enhanced FireWorks Algorithm (EFWA), which is an improved version of FWA. The authors use the MapReduce platform for parallelizing EFWA, investigate its ability to scale, and report on the speedup obtained on different benchmark functions for increasing problem dimensions.

Video based vehicle classification is gaining huge grounds due to its low cost and satisfactory accuracy. This paper presents a robust vehicle classification system. The system in its essence, aims to classify a vehicle based on the number of circles (axles) in an image using Hough Transform which is a popular parameter based feature detection method. The system consists of four modules whereby the output of one module feeds the next in line. We test our system on single lane highway and street traffic. When the information about the problem at hand (changing weather conditions, camera calibration parameters etc.) is limited or is dynamic, determining the Hough Transform set-up parameters manually becomes time consuming, challenging, and may often lead to false detections. This calls for finding the appropriate parameter-set dynamically according to the situation, which inherently is a global optimization problem. Differential Evolution has emerged as a simple and efficient global optimizer, and we couple it with Hough Transform to improve the overall accuracy of the classification system. We test five different variants of DE on varied videos, and provide a performance profile of all the variants. Our results demonstrate that employing DE indeed improves the system's classification accuracy (at the expense of extra compute cycles) making the system more reliable and robust.