PSO optimized SVM parameters for far infrared pedestrian detection

Aiming at the problem that the uncertainty of manual selection of penalty factor C and Gauss kernel parameter γ of support vector machine (SVM) in OpenCV leads to the unsatisfactory accuracy of infrared pedestrian detection, an infrared pedestrian detection method based on particle swarm optimization (PSO) optimized SVM is proposed. Samples are selected to establish the sample database. HOG feature vectors are extracted from the samples to calculate the feature matrix and put into SVM for training. Then, PSO is used to optimize the parameters of penalty factor and Gauss kernel, and SVM is trained again to get the best pedestrian classifier model, which is used to identify pedestrians. The results show that by applying PSO optimized SVM parameters to far-infrared pedestrian detection, the rate of missed detection and false detection is significantly reduced, the accuracy of pedestrian classification is significantly improved, and the operating time is shortened. © 2019, Politechnica University of Bucharest. All rights reserved.