Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Vehicle Classification
The technologies for automated vehicle classification have been evolving over decades. With rapidly growing affordable sensors such as closed‐circuit television (CCTV) cameras, light detection and ranging (LiDAR), and even thermal imaging devices, we are able to detect, track, and categorize vehicles in multiple lanes simultaneously. In this chapter, we focus on movable sensors, including RGB video cameras, thermal imaging sensors, and LiDAR imaging sensors. The different categories of algorithms include heuristics and statistics‐based, shape‐based, and feature‐based methods.
The shape‐based vehicle classification methods, such as histogram of oriented gradient (HOG) and Haar‐like features, require substantial training images and carefully designed preprocessing algorithms, such as deformable model, sample light and size normalization, and sample rotation resampling. The challenge emerges when multiple vehicles are clustered with others in the field of view of the camera. The silhouette‐based methods such as vehicle contours are one of the simpler classification methods. However, they need relatively clean blobs to start with. For stationary CCTV cameras, this kind of approach is feasible using background segmentation methods. Simple methods such as proportion‐based modeling were successful methods in our highway case studies. However, the amount and variety of vehicles it can handle is very limited due to the lack of proportion specifications for every vehicle type.
In many cases, we can also utilize a hierarchical classification, which allows us to divide the classification process into a coarse and a fine stage. The former involves a rough filtering based on silhouette measurement, while the latter stage is responsible for further distinction between objects in more similar subclasses. The accuracy of vehicle classification can be improved by combining multiple sensors such as thermal imaging, LiDAR imaging, and RGB visible cameras. When possible, we can even incorporate acoustic sound signals to detect unique vehicle signatures, such as distinctive sounds made by motorcycles and trucks.
Vehicle Classification
The technologies for automated vehicle classification have been evolving over decades. With rapidly growing affordable sensors such as closed‐circuit television (CCTV) cameras, light detection and ranging (LiDAR), and even thermal imaging devices, we are able to detect, track, and categorize vehicles in multiple lanes simultaneously. In this chapter, we focus on movable sensors, including RGB video cameras, thermal imaging sensors, and LiDAR imaging sensors. The different categories of algorithms include heuristics and statistics‐based, shape‐based, and feature‐based methods.
The shape‐based vehicle classification methods, such as histogram of oriented gradient (HOG) and Haar‐like features, require substantial training images and carefully designed preprocessing algorithms, such as deformable model, sample light and size normalization, and sample rotation resampling. The challenge emerges when multiple vehicles are clustered with others in the field of view of the camera. The silhouette‐based methods such as vehicle contours are one of the simpler classification methods. However, they need relatively clean blobs to start with. For stationary CCTV cameras, this kind of approach is feasible using background segmentation methods. Simple methods such as proportion‐based modeling were successful methods in our highway case studies. However, the amount and variety of vehicles it can handle is very limited due to the lack of proportion specifications for every vehicle type.
In many cases, we can also utilize a hierarchical classification, which allows us to divide the classification process into a coarse and a fine stage. The former involves a rough filtering based on silhouette measurement, while the latter stage is responsible for further distinction between objects in more similar subclasses. The accuracy of vehicle classification can be improved by combining multiple sensors such as thermal imaging, LiDAR imaging, and RGB visible cameras. When possible, we can even incorporate acoustic sound signals to detect unique vehicle signatures, such as distinctive sounds made by motorcycles and trucks.
Vehicle Classification
Loce, Robert P. (Herausgeber:in) / Bala, Raja (Herausgeber:in) / Trivedi, Mohan (Herausgeber:in) / Deshpande, Shashank (Autor:in) / Muron, Wiktor (Autor:in) / Cai, Yang (Autor:in)
14.03.2017
33 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
LiDAR imaging sensors , 3D model‐based , k‐NN , AVC , Haar‐like features , histogram of oriented gradients (HOG) , eigenvehicles , background subtraction , PCA , shape feature descriptors , thermal imaging , edge‐based , LDA , intrinsic proportion model , transformation , shape‐based , vehicle classification , fusion , video‐based classification , SIFT , profile‐based , hierarchical classification , vehicle detection , thermal signature , LiDAR , edge map matching , UAV , SVM
Vehicle Classification Using Advanced Technologies
| British Library Online Contents | 1996
Amphibious garbage collection and classification vehicle
| Europäisches Patentamt | 2020
Side-Fire Lidar-Based Vehicle Classification
| British Library Online Contents | 2012
Moving Vehicle Classification Using Cloud Model
| British Library Online Contents | 2011