D. ClassificationThe classification of images is the most challenging

D. ClassificationThe classification of images is the most challenging taskfor the automatic detection of motorcycles from traffic videos.Classification might provide the answer whether the movingobjects in a traffic video is motorcycle or non-motorcycle. It isa type of supervised machine learning, where an algorithmlearns to classify new observations from examples of labeleddata. For classification purposes Support Vector Machine(SVM) algorithm, have been used. SVM is an approach todata classification that estimates how likely a data point is tobe a member of one group or the other depending on thetraining dataset.TrainingThe entire process of training the SVM classifier, the 2classes based on the hyperplane is called training. For trainingthe system, the MATLAB function svmtrain is used as shownbelow.model = svmtrain(traininglabelvector, traininginstancematrix’libsvmoptions’)TestingIn this stage the trained model is used to test the featurevector in order to predict the value. This process recognizesthe objects from the labels and features represented by thetrained model. The MATLAB function svmpredict used fortesting.predictedlabel, accuracy, decisionvalues/probestimates =svmpredict(testinglabelvector,testinginstancematrix, model ‘libsvmoptions’)IV. RESULT & DISCUSSIONSThe proposed methodologies were implemented withMATLAB R2014a, on traffic videos taken from CCTVcameras. The system machine was standard desktop withoutany specific hardware or software optimization. The imagedatabase contains 2906 images of non-motorcycles and 1002images of motorcycles. Fig 7-8 show a sequence of imagesfrom the output of our proposed motorcycle recognitionsystem.Fig 7. Multiple non-motorcycle detection on highway roadFig 8. Both motorcycle and non-motorcycle recognitionThe results of the algorithm in various videos are shown inTable 1.Table 1. Quantitative evaluation of classification method on aset of 7 microscopic videos.The proposed method have an accuracy rate of 0.9523 formoving object segmentation and 0.9119 for motorcyclerecognition. The algorithm performance is evaluated based onTrue Positive (TP), False Positive (FP), False Negative (FN),True Negative (TN) and accuracy 8. Fig 9. shows theReceiver Operating Characteristic (ROC) 9 of the proposedapproach. This curve reports the true positive rate against thefalse positive rate. This figure confirms that the proposedapproach is suitable for our task in order to identifymotorcycles in traffic videos.Fig 9. ROC curve of the proposed method


I'm Elaine!

Would you like to get a custom essay? How about receiving a customized one?

Check it out