SIFT-BASED MEASUREMENTS FOR VEHICLE MODEL RECOGNITION

摘要: Abstract – A SIFT-based Vehicle Manufacturer and Model Recognition (VMMR) method was utilized to tackle the problem of vehicle security. Distinctive parts frontal view such as headlights, grill logo area were segmented. series experiments conducted in a variety outdoor conditions, where query image that rotated, scaled, shifted or set different lighting matched against database model images. In this work, is shown processing functions based on Scale Invariant Feature Transform (SIFT) measurements can be used obtain high performance object features recognition, creating key-point fingerprint (pattern) for each class. majority cases, SIFT performs very well, terms efficiency robustness . Keywords : vehicle, measurement, 1. INTRODUCTION Image matching fundamental computer vision which occurs many applications from fields including retrieval security enforcement robot navigation. Content Based Retrieval (CBIR) addresses images sharing similar visual content common approach accurate known “keypoint” “interesting point” extraction comparison. It involves identifying points reliably extracted same category objects. Earlier research into invariant keypoints focused invariance rotation translation, Siggelkow [1], Schultz-Mirbach [2]. Transforms introduced by Lowe [3], [4], [5] they are rotation, translation scale variation between partially affine distortion, illumination variance noise. Research related fully features, published Brown [6], Mikolajczyk Schmidt [7].Vehicle classification general categories task has been adequately addressed literature Weber [8], Kato [9], Lai [10], [11]. Approaches with identification have previously encouraging results. Dlangenkov Belongie [12] making them suitable Recognition, using rear-view Petrovic Cootes [13] presented an interesting recognition verification displays best results respective tasks. Merler [14] presents car detection system color segmentation labeling, recognition. Conos [15] deals type He proposed descriptor feature but his computational expensive -in some cases takes more than 12 hours accomplished. novel proposed, whose aim reliable manufacturer model, (eg. Alfa Romeo 156), models. This effort assisted developed license plate module Anagnostopoulos [16], C. Anagnostopoulos, I. Loumos Kayafas [17] special technique, called phase congruency Covesi [18], Psyllos, [19], [20]. The consists mainly six modules: 1) License Plate 2) Frontal View Segmentation, 3) Mask 4) Matching, 5) 6) depicted Figure