2026-06-11T00:14:51.586Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:1:9867 2024-05-25T08:05:50Z 1

Firefly-Xgboost Approach for Pedestrian Detection 2022 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/1/9867 https://ieeexplore.ieee.org/document/9840700 A. Petrovic I. Strumberger M. Antonijevic D. Jovanovic D. Mladenovic A. Chhabra The area of machine learning, particularly computer vision, has many applications in various fields, including defect detection, metrology, intrusion detection, autonomous vehicles, etc. However, one of the most important challenge from this domain is that each model should be adapted for specific dataset. This problem is known as the hyperparameters optimization and refers to tuning untrainable model parameters for the specific task. Research proposed in this paper introduces hybrid framework between machine learning and swarm intelligence for pedestrian detection, which is very important for autonomous vehicles. With the goal of improving classification accuracy, the histogram of oriented gradients and local binary pattern features are first used to describe pedestrians and further these features are used as the input for the extreme gradient boosting classifier. During the optimization process, the eXtreme gradient boosting hyperparameters were tuned by using well-known firefly algorithm swarm intelligence metaheuristics. Proposed hybrid framework was validated against INRIA person datasets and achieved results were compared to those obtained by the support vector machine, as well as genetic algorithm method that was also utilized for eXtreme gradiend boosting tuning. Comparative analysis results show that the firefly algorithm has huge potential in this domain. conferenceObject 197 202 10.1109/ZINC55034.2022.9840700 2022 IEEE Zooming Innovation in Consumer Technologies Conference (ZINC), Novi Sad, Serbia