Table 1.

Related work summary.

Reference	Input	Method	Contribution
[6,8,9,10,14]	Single-view	ML	Hand-crafted geometric features represent vehicles for detection and classification using ML-based algorithms
[11,12]	Single-view	DL	CNN models are proposed to perform automatic feature learning for vehicle detection and classification
[65]	Single-view	Eigenvalue decomposition	Eigenvehicles are introduced as an unsupervised feature representation method for vehicle recognition
[66,67,68]	Single-view	Nonnegative factorization	A part-based model is employed for vehicle recognition via non-negative matrix/tensor factorization
[72,73,74]	Single-view	DL-based MTL	MTL models based on DL are employed to simultaneously perform multiple tasks, including road segmentation, vehicle detection and classification
[92]	Multi-view	DL	This work employs a YOLO-based model that fuses camera and LiDAR data at multiple levels
[61,93,94]	Single-view	ML	Single-view features, such as HOG, Haar wavelets, or GLCM, represent vehicles for classification in ML models
[95]	Multi-view	Tucker decomposition	A tensor decomposition is employed for feature selection of HOG, LBP, and FDF features
[70,71,96]	Multi-view	MVL	MVL approaches are proposed to enhance vehicle detection, classification, and background modeling by learning richer data representations from color features
[30,97,98,99,100]	−	−	These works provide theoretical foundations on tensors and its operations, such as the Einstein and Hadamard products, with applications across multiple domains
[32,77,78,79,80,81,82,83,90]	−	DL	Matrix and tensor decompositions are employed for speeding up CNNs by compressing FC and Conv layers and reducing the dimensionality of their input space
[91]	−	DL	Multilinear layers are introduced for dimensionality reduction and regression purposes in CNNs, leveraging tensor decompositions for efficient computation.