Vehicle-Classification Algorithm for Single-Loop Detectors Using Neural Networks
TL;DR: A new algorithm for ILD using back-propagation neural networks is suggested to improve the vehicle-classification accuracy compared to the conventional method based on ILD.
Abstract: Vehicle class is an important parameter in the process of road-traffic measurement. Currently, inductive-loop detectors (ILD) and image sensors are rarely used for vehicle classification because of their low accuracy. To improve the accuracy, the authors suggest a new algorithm for ILD using back-propagation neural networks. In the developed algorithm, the inputs to the neural networks are the variation rate of frequency and frequency waveform. The output is five classified vehicles. The developed algorithm was assessed at test sites, and the recognition rate was 91.5%. The results verified that the proposed algorithm improves the vehicle-classification accuracy compared to the conventional method based on ILD
Citations
More filters
TL;DR: This work adds a regularization term to the cross-entropy loss and proposes a new loss function, Dual Cross-Entropy Loss, which improves the fine-grained vehicle classification performance and has good performance on three other general image classification tasks.
Abstract: Fine-grained vehicle classification is a challenging topic in computer vision due to the high intraclass variance and low interclass variance. Recently, considerable progress has been made in fine-grained vehicle classification due to the huge success of deep neural networks. Most studies of fine-grained vehicle classification based on neural networks, focus on the neural network structure to improve the classification performance. In contrast to existing works on fine-grained vehicle classification, we focus on the loss function of the neural network. We add a regularization term to the cross-entropy loss and propose a new loss function, Dual Cross-Entropy Loss . The regularization term places a constraint on the probability that a data point is assigned to a class other than its ground-truth class, which can alleviate the vanishing of the gradient when the value of the cross-entropy loss is close to zero. To demonstrate the effectiveness of our loss function, we perform two sets of experiments. The first set is conducted on a small-sample fine-grained vehicle classification dataset, the Stanford Cars-196 dataset. The second set is conducted on two small-sample datasets, the LabelMe dataset and the UIUC-Sports dataset, as well as on one large-sample dataset, the CIFAR-10 dataset. The experimental results show that the proposed loss function improves the fine-grained vehicle classification performance and has good performance on three other general image classification tasks.
105 citations
Cites background from "Vehicle-Classification Algorithm fo..."
...Research on vehicles has received considerable attention [1]–[3], including applications in the field of computer vision, such as vehicle classification [4]–[7], vehicle detection [8]–[10], vehicle segmentation [11], vehicle re-identification (re-ID) [12], [13], and fine-grained vehicle classification [14], [15]....
[...]
TL;DR: A new sensing device that can simultaneously monitor traffic congestion and urban flash floods is presented, based on the combination of passive infrared sensors and ultrasonic rangefinder, and relies on dynamic Bayesian Networks to fuse heterogeneous data both spatially and temporally for vehicle detection.
Abstract: In this paper, a new sensing device that can simultaneously monitor traffic congestion and urban flash floods is presented. This sensing device is based on the combination of passive infrared sensors (PIRs) and ultrasonic rangefinder, and is used for real-time vehicle detection, classification, and speed estimation in the context of wireless sensor networks. This framework relies on dynamic Bayesian Networks to fuse heterogeneous data both spatially and temporally for vehicle detection. To estimate the speed of the incoming vehicles, we first use cross correlation and wavelet transform-based methods to estimate the time delay between the signals of different sensors. We then propose a calibration and self-correction model based on Bayesian Networks to make a joint inference by all sensors about the speed and the length of the detected vehicle. Furthermore, we use the measurements of the ultrasonic and the PIR sensors to perform vehicle classification. Validation data (using an experimental dual infrared and ultrasonic traffic sensor) show a 99% accuracy in vehicle detection, a mean error of 5 kph in vehicle speed estimation, a mean error of 0.7m in vehicle length estimation, and a high accuracy in vehicle classification. Finally, we discuss the computational performance of the algorithm, and show that this framework can be implemented on low-power computational devices within a wireless sensor network setting. Such decentralized processing greatly improves the energy consumption of the system and minimizes bandwidth usage.
101 citations
Cites background from "Vehicle-Classification Algorithm fo..."
...Other types of sensors that are used for vehicle classification include inductive loops [13], [26], magnetic sensors [25], [5], [12], weigh-in-motion, piezoelectric cables, and pneumatic tubes....
[...]
TL;DR: A novel inductive loop sensor that can detect vehicles under a heterogeneous and less-lane-disciplined traffic and thus can be used to support a traffic control management system in optimizing the best use of existing roads is presented.
Abstract: This paper presents a novel inductive loop sensor that can detect vehicles under a heterogeneous and less-lane-disciplined traffic and thus can be used to support a traffic control management system in optimizing the best use of existing roads. The loop sensor proposed in this paper detects large (e.g., bus) as well as small (e.g., bicycle) vehicles occupying any available space in the roadway, which is the main requirement for sensing heterogeneous and lane-less traffic. To accomplish the sensing of large as well as small vehicles, a multiple loop system with a new inductive loop sensor structure is proposed. The proposed sensor structure not only senses and segregates the vehicle type as bicycle, motor cycle, scooter, car, and bus but also enables accurate counting of the number of vehicles even in a mixed traffic flow condition. A prototype of the multiple loop sensing system has been developed and tested. Field tests indicate that the prototype successfully detected all types of vehicles and counted, correctly, the number of each type of vehicles. Thus, the suitability of the proposed sensor system for any type of traffic has been established.
92 citations
Cites background from "Vehicle-Classification Algorithm fo..."
...Research papers discussing improvement of loop detectors for better speed measurement [8], [9] and classification [10]–[16] for lane-disciplined traffic conditions were reported....
[...]
TL;DR: A novel vehicle-classification algorithm that uses the time-variable signal generated by a single inductive loop detector to strengthen the classification accuracy by emphasizing the undercarriage height variation of the vehicle.
Abstract: This paper presents a novel vehicle-classification algorithm that uses the time-variable signal generated by a single inductive loop detector In earlier studies, the noisy raw signal was fed into the algorithm by reducing its size with rough sampling However, this approach loses the original signal form and cannot be the best exemplar vector The developed algorithm suggests three contributions to cope with these problems The first contribution is to clear the noise with discrete Fourier transform (DFT) The second contribution is to transfer the noiseless pattern into the Principal Component Analysis (PCA) domain PCA is exploited not only for decorrelation but for explicit dimensionality reduction as well This goal cannot be achieved by simple raw data sampling The last contribution is to expand the principal components with a local maximum (Lmax) parameter It strengthens the classification accuracy by emphasizing the undercarriage height variation of the vehicle These parameters are fed into the three-layered backpropagation neural network (BPNN) BPNN classifies the vehicles into five groups, and the recognition rate is 9421% This recognition rate has performed best, compared with the methods presented in published works
67 citations
Cites background or methods from "Vehicle-Classification Algorithm fo..."
...However, its accuracy is low [3], [4], and its installation has difficulties (i....
[...]
...To assess the algorithm presented in this paper, the novel method has been added to the vehicle classification comparison table published in Ki’s study [4]....
[...]
...1) is the most common way of understanding a vehicle’s presence on highways through the measurement of frequency variation [4]....
[...]
...1) Environmental Conditions: ILs are placed under the road surface and are connected to the control unit by a lead in the wire....
[...]
...The artificial neural network (ANN)-based pattern recognition approaches have been used to improve the accuracy of vehicle classification on ILs [4]....
[...]
TL;DR: The results of the case study indicate that ITS technologies can be a useful tool for real-time estimations of CO2 emissions with a high spatiotemporal resolution.
Abstract: In this paper, a bottom-up vehicle emission model is proposed to estimate real-time CO2 emissions using intelligent transportation system (ITS) technologies. In the proposed model, traffic data that were collected by ITS are fully utilized to estimate detailed vehicle technology data (e.g., vehicle type) and driving pattern data (e.g., speed, acceleration, and road slope) in the road network. The road network is divided into a set of small road segments to consider the effects of heterogeneous speeds within a road link. A real-world case study in Beijing, China, is carried out to demonstrate the applicability of the proposed model. The spatiotemporal distributions of CO2 emissions in Beijing are analyzed and discussed. The results of the case study indicate that ITS technologies can be a useful tool for real-time estimations of CO2 emissions with a high spatiotemporal resolution.
62 citations
References
More filters
TL;DR: Algorithm for vision-based detection and classification of vehicles in monocular image sequences of traffic scenes recorded by a stationary camera based on the establishment of correspondences between regions and vehicles, as the vehicles move through the image sequence is presented.
Abstract: This paper presents algorithms for vision-based detection and classification of vehicles in monocular image sequences of traffic scenes recorded by a stationary camera. Processing is done at three levels: raw images, region level, and vehicle level. Vehicles are modeled as rectangular patches with certain dynamic behavior. The proposed method is based on the establishment of correspondences between regions and vehicles, as the vehicles move through the image sequence. Experimental results from highway scenes are provided which demonstrate the effectiveness of the method. We also briefly describe an interactive camera calibration tool that we have developed for recovering the camera parameters using features in the image selected by the user.
833 citations
21 May 2001
TL;DR: In this article, the influence of loop length (in direction of vehicle movement) on differences between characteristics describing the magnetic profiles of the vehicles belonging to the different classes is discussed, and the case of extremely short loop (10 cm) which allows detection of the number of axles is also analyzed.
Abstract: The class of vehicle is one of more important parameters in the process of road traffic measurement. Up to now, strip piezoelectric sensors and video systems have been used. The use of very cheap inductive loop detectors for vehicle classification is also possible. Such vehicle classification systems are based on magnetic profiles recorded from inductive loops. The magnetic profile is sensitive to the loop dimensions. This paper presents a discussion concerning the influence of loop length (in direction of vehicle movement) on differences between characteristics describing the magnetic profiles of the vehicles belonging to the different classes. As characteristics describing the magnetic profile of the vehicle have been used: magnetic profiles in time domain (normalized in amplitude), probability density function and magnetic profiles in vehicle length domain. For real time applications, the conversion of the measured signal into a vector of numerical parameters (a few only) is also proposed. The influence of loop dimensions on a chosen signal parameter was investigated. The case of extremely short loop (10 cm), which allows detection of the number of axles, was also analyzed.
148 citations
"Vehicle-Classification Algorithm fo..." refers background in this paper
...Such a signal is known as the magnetic profile of the vehicle and is the base of the vehicleclassification data-processing algorithm [1]....
[...]
...Vehicle class is important for access control (areas closed for some vehicle classes, limits of velocity, and weight for different vehicle classes), statistic purposes, and weighing of vehicles in motion [1]–[3]....
[...]
TL;DR: In this paper, a formal methodology for testing speed traps off-line has been developed, and ways to extend the work to on-line testing are suggested, and the work is used to evaluate several loop sensor units, revealing problems in two models.
Abstract: Dual loop speed traps have a distinct advantage over single loop detectors because the speed trap detection system is redundant. Each vehicle is observed twice under normal operating conditions, once at each loop. The two observations are normally used to measure velocity, but as this paper demonstrates, the redundancy can also be used to assess the performance of the speed trap and identify detector errors. At free-flow velocities, the time each detector is occupied by a vehicle (i.e., the on-time) should be virtually identical, regardless of the vehicle length. Many hardware errors will cause the two on-times to differ. Exploiting this property, a formal methodology for testing speed traps off-line has been developed, and ways to extend the work to on-line testing are suggested. The work is used to evaluate several loop sensor units, revealing problems in two models. A second example shows how the work can be used to detect cross talk between sensor units.
137 citations
"Vehicle-Classification Algorithm fo..." refers background in this paper
...Since its introduction in the early 1960s, the inductive-loop detector (ILD) has become the most popular form of detection system [4], [5], and most traffic-surveillance applications depend on it....
[...]
TL;DR: A new model that uses an error-filtering algorithm to improve the accuracy of speed measurements and it can be concluded that the proposed model significantly improves vehicle-speed-measuring accuracy.
Abstract: Vehicle speed is an important parameter in measurements of road traffic. At present, double-loop detectors are generally used for vehicular speed measurement. However, these detectors incur errors caused by scanning time, spacing between double loops, irregular vehicle trajectories, and the presence of multiple vehicles in the detection zone. This paper suggests a new model that uses an error-filtering algorithm to improve the accuracy of speed measurements. In the field tests, all percent errors of the vehicular speeds measured by the proposed model were within the error tolerance limit (plusmn5%). Furthermore, the variance of percent errors was reduced. Therefore, it can be concluded that the proposed model significantly improves vehicle-speed-measuring accuracy
89 citations
"Vehicle-Classification Algorithm fo..." refers methods in this paper
...For several years, we have conducted speed-accuracy tests for automated speed-enforcement systems using loop detectors [ 20 ]....
[...]
06 May 2001
TL;DR: Experimental results show that the parameterized model can satisfactorily and effectively describe vehicles, and the correct rate for vehicle recognition using neural networks classifier is more than 91%.
Abstract: This paper presents a novel method of vehicle classification using parameterized model and neural networks. First, we propose the parameterized model, which can describe the features of vehicle. In this model, vertices and their topological structure are regarded as the key features. Then we adopt a classifier based on multi-layer perceptron networks (MLPN) to recognize vehicles. In this neural network classifier, learning algorithms based on the gradient descent method for the least exponential function error (LEFE) are adopted. Experimental results show that the parameterized model can satisfactorily and effectively describe vehicles, and the correct rate for vehicle recognition using neural networks classifier is more than 91%. This novel method can be used in real world systems such as the vehicle verifying system in toll collecting station. However, it is not difficult to adapt algorithms and improve the model to fit for other traffic scene.
73 citations