Conference material: "Proceedings of the International Conference on Computer Graphics and Vision “Graphicon” (19-21 September 2022, Ryazan)"
Authors:Strotov V.V., Zhgutov P.E.
Combining Several Algorithmsto Increase the Accuracy of Pedestrian Detection and Localization
Abstract:
In this paper, the problem of developing an algorithm for detecting pedestrians in a sequence of images is considered. The concept of a combined algorithm combining a pedestrian detector with an aggregation function of several channels, together with a motion detection algorithm and a tracking function is proposed. The paper describes the stages of the algorithm, as well as experimental studies that were conducted on a set of test videos. During the study of the combined algorithm, the quality of detection of the combined algorithm and the pedestrian detector was compared separately. The study of the dependence of the results of the algorithms on changes in the detector parameters was carried out. Based on the results of experimental studies, a conclusion was made about the best indicators of accuracy and quality of detection in a combined algorithm that can find application in computer vision systems.
Keywords:
people detector with aggregate channel feature (ACF), a method for calculating features in a multiscale representation of a signal (Fast Feature Pyramids), formatting examples, move extract method, track function, region of interesting (ROI)
