Mid-level-Representation Based Lexicon for Vehicle Make and Model Recognition

Real-Time Vehicle Make and Model Recognition Based on a Bag of SURF Features

“…In contrast to their work, we investigate optimal dictionary building parameters in the context of VMMR challenges, through two schemes of dictionary building. Amongst the most recent works on VMMR is that of Fraz et al [12]. They form a lexicon that is comprised of all training images' features as words.…”

“…Their MLR construction is computationally expensive, reported to consume about 0.4 s per image, and hence unsuitable for real-time VMMR. Unlike [12], we learn a dictionary by retaining only the dominant features of training images as codewords, and not all the features.…”

“…Other features based on edges, gradients or corners (e.g., by [2], [3], [16]), and MPEG-7 descriptors such as Edge Histograms [17], [18] (e.g., by [10]) were also explored for VMMR purposes. In most approaches, the raw features are embedded into global representations of vehicle makes and models ( [3], [5], [10], [12], [15]) as shown in Table I. Some works (such as [12]) refer to the global representations as Mid Level Representation (MLR).…”

“…In most approaches, the raw features are embedded into global representations of vehicle makes and models ( [3], [5], [10], [12], [15]) as shown in Table I. Some works (such as [12]) refer to the global representations as Mid Level Representation (MLR). The quality of a global features representation technique is assessed by its processing speed, computational complexity in forming the holistic representations, and VMMR accuracy, which reflects its discriminative power in representing the different makes and models while generalizing over the multiplicity issues within a make-model class.…”

“…Works such as [4] use raw image features like Scale Invariant Feature Transform (SIFT [9]) to describe make-model instances. In fact, SIFT has been used by many VMMR works such as [4], [5], [10]- [12]. Due to the high dimensionality and relatively slow computational speed of SIFT, some works have adopted the Speeded Up Robust Features (SURF [13]) (e.g., [5], [10]) and the Histogram of Oriented Gradients (HOG [14]) (e.g., in [5] and [15]).…”