Maghsood Salimi, Mohammad Loni, M. Sirjani, A. Cicchetti, Sara Abbaspour Asadollah
Convolutional Neural Networks (CNNs) have received great attention in the computer vision domain. However, CNNs are vulnerable to adversarial attacks, which are manipulations of input data that are imperceptible to humans but can fool the network. Several studies tried to address this issue, which can be divided into two categories: (i) training the network with adversarial examples, and (ii) optimizing the network architecture and/or hyperparameters. Although adversarial training is a sufficient defense mechanism, they suffer from requiring a large volume of training samples to cover a wide perturbation bound. Tweaking network activation functions (AFs) has been shown to provide promising results where CNNs suffer from performance loss. However, optimizing network AFs for compensating the negative impacts of adversarial attacks has not been addressed in the literature. This paper proposes the idea of searching for AFs that are robust against adversarial attacks. To this aim, we leverage the Simulated Annealing (SA) algorithm with a fast convergence time. This proposed method is called SARAF. We demonstrate the consistent effectiveness of SARAF by achieving up to 16.92%, 18.3%, and 15.57% accuracy improvement against BIM, FGSM, and PGD adversarial attacks, respectively, over ResNet-18 with ReLU AFs (baseline) trained on CIFAR-10. Meanwhile, SARAF provides a significant search efficiency compared to random search as the optimization baseline.
{"title":"SARAF: Searching for Adversarial Robust Activation Functions","authors":"Maghsood Salimi, Mohammad Loni, M. Sirjani, A. Cicchetti, Sara Abbaspour Asadollah","doi":"10.1145/3589572.3589598","DOIUrl":"https://doi.org/10.1145/3589572.3589598","url":null,"abstract":"Convolutional Neural Networks (CNNs) have received great attention in the computer vision domain. However, CNNs are vulnerable to adversarial attacks, which are manipulations of input data that are imperceptible to humans but can fool the network. Several studies tried to address this issue, which can be divided into two categories: (i) training the network with adversarial examples, and (ii) optimizing the network architecture and/or hyperparameters. Although adversarial training is a sufficient defense mechanism, they suffer from requiring a large volume of training samples to cover a wide perturbation bound. Tweaking network activation functions (AFs) has been shown to provide promising results where CNNs suffer from performance loss. However, optimizing network AFs for compensating the negative impacts of adversarial attacks has not been addressed in the literature. This paper proposes the idea of searching for AFs that are robust against adversarial attacks. To this aim, we leverage the Simulated Annealing (SA) algorithm with a fast convergence time. This proposed method is called SARAF. We demonstrate the consistent effectiveness of SARAF by achieving up to 16.92%, 18.3%, and 15.57% accuracy improvement against BIM, FGSM, and PGD adversarial attacks, respectively, over ResNet-18 with ReLU AFs (baseline) trained on CIFAR-10. Meanwhile, SARAF provides a significant search efficiency compared to random search as the optimization baseline.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127493308","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Customs clearance is a challenging and time-consuming process that must be completed in the sphere of international trade. As a result, the cargo is frequently delayed at the port. If the personnel know the initial number of items, they may be able to continue with other procedures even when they are not physically present at the location. Image processing is helpful in this area since it allows for the prediction of the type of goods based on the appearance of the package. This allows for the determination of the quantity of each type of product prior to the arrival of the employees at the site. Three distinct import-export companies contributed 5,675 photos, and a machine learning approach was used to create a model that can predict the types of things that fall into one of five categories. Also, the researchers made an OCR-based classification algorithm with the goal of making machine learning work better for certain types of things that have trouble learning.
{"title":"Employing Machine Learning and an OCR Validation Technique to Identify Product Category Based on Visible Packaging Features","authors":"Takorn Prexawanprasut, Lalita Santiworarak, Piyaporn Nurarak, Poom Juasiripukdee","doi":"10.1145/3589572.3589589","DOIUrl":"https://doi.org/10.1145/3589572.3589589","url":null,"abstract":"Customs clearance is a challenging and time-consuming process that must be completed in the sphere of international trade. As a result, the cargo is frequently delayed at the port. If the personnel know the initial number of items, they may be able to continue with other procedures even when they are not physically present at the location. Image processing is helpful in this area since it allows for the prediction of the type of goods based on the appearance of the package. This allows for the determination of the quantity of each type of product prior to the arrival of the employees at the site. Three distinct import-export companies contributed 5,675 photos, and a machine learning approach was used to create a model that can predict the types of things that fall into one of five categories. Also, the researchers made an OCR-based classification algorithm with the goal of making machine learning work better for certain types of things that have trouble learning.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"64 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125101072","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Bin Yi, Wenqiang Lin, Wenqi Li, Xiaohua Gao, Bing Zhou, Jun Tang
In the view of the existing workshop process quality prediction method for the process parameters related timing information mining is not sufficient, existing research does not consider the contribution of different characteristics to the prediction target difference, this paper proposes the fusion of attention mechanism, convolutional neural network and time convolutional network. The attention module adaptively allocates weight information to the input features, convolutional neural network module to deeply mine the correlation information between process parameters was used, extracts the temporal information between process sequences with time convolutional neural learning, and finally superposition the full connection network mapping to obtain the workshop process quality prediction value. After example verification, the experimental results show that the constructed model is better than other process quality prediction models in the prediction accuracy, stability and network structure.
{"title":"Process Quality Prediction Algorithm of Multi output Workshop Based on ATT-CNN-TCN","authors":"Bin Yi, Wenqiang Lin, Wenqi Li, Xiaohua Gao, Bing Zhou, Jun Tang","doi":"10.1145/3589572.3589590","DOIUrl":"https://doi.org/10.1145/3589572.3589590","url":null,"abstract":"In the view of the existing workshop process quality prediction method for the process parameters related timing information mining is not sufficient, existing research does not consider the contribution of different characteristics to the prediction target difference, this paper proposes the fusion of attention mechanism, convolutional neural network and time convolutional network. The attention module adaptively allocates weight information to the input features, convolutional neural network module to deeply mine the correlation information between process parameters was used, extracts the temporal information between process sequences with time convolutional neural learning, and finally superposition the full connection network mapping to obtain the workshop process quality prediction value. After example verification, the experimental results show that the constructed model is better than other process quality prediction models in the prediction accuracy, stability and network structure.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128650230","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Kernel method is widely used in machine learning. At present, the connection between kernel methods and quantum computing has been gradually established, which provides a new algorithm idea for the field of quantum machine learning. Research shows that the construction of minimized quantum circuits can be reliably performed on Noisy Intermediate-Scale Quantum (NISQ) devices. This paper proposes a compact quantum classifier based on kernel method. By introducing the compact amplitude encoding, the data label of the phase corresponding to the quantum state is encoded. Compared with the proposed classifier based on quantum kernel method, it can reduce 2 quantum registers, further reduce the circuit depth, and thus reduce the algorithm complexity. The double qubit measurement is simplified to single qubit measurement. In addition, this model achieves the optimal variance in quantum circuit parameters, which can effectively save computational resources. Experimental simulation shows that the expected value measurement in the proposed classifier model is closer to the theoretical value, and the classification accuracy is more accurate. At the same time, the system model has low entanglement, which can effectively reduce the cost of the whole preparation.
{"title":"Research on Compact Quantum Classifier Based on Kernel Method","authors":"Ruihong Jia, Guang Yang, Min Nie, Yun Zhang","doi":"10.1145/3589572.3589592","DOIUrl":"https://doi.org/10.1145/3589572.3589592","url":null,"abstract":"Kernel method is widely used in machine learning. At present, the connection between kernel methods and quantum computing has been gradually established, which provides a new algorithm idea for the field of quantum machine learning. Research shows that the construction of minimized quantum circuits can be reliably performed on Noisy Intermediate-Scale Quantum (NISQ) devices. This paper proposes a compact quantum classifier based on kernel method. By introducing the compact amplitude encoding, the data label of the phase corresponding to the quantum state is encoded. Compared with the proposed classifier based on quantum kernel method, it can reduce 2 quantum registers, further reduce the circuit depth, and thus reduce the algorithm complexity. The double qubit measurement is simplified to single qubit measurement. In addition, this model achieves the optimal variance in quantum circuit parameters, which can effectively save computational resources. Experimental simulation shows that the expected value measurement in the proposed classifier model is closer to the theoretical value, and the classification accuracy is more accurate. At the same time, the system model has low entanglement, which can effectively reduce the cost of the whole preparation.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129147980","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As robots become more ubiquitous in our daily life, it has become very important to extract task and environmental information through more natural, meaningful, and easy-to-use interaction interfaces. Not only this helps the user to adapt to (thus trust) a robot in a collaborative environment, it can supplement the core sensory information, helping the robot make reliable decisions. This paper presents a framework that combines two natural interaction interfaces: speech and gaze to reliably infer the object of interest and the robotic task parameters. The gaze estimation module utilizes pre-defined 3D facial points and matches them to a set of extracted estimated 3D facial landmarks of the users from 2D images to infer the gaze direction. Subsequently, the verbal instructions are passed through a deep learning model to extract the information relevant to a robotic task. These extracted task parameters from verbal instructions along with the estimated gaze directions are combined to detect and/or disambiguate objects in the scene to generate the final task configurations. The proposed framework shows very promising results in integrating the relevant task parameters for the intended robotic tasks in different real-world interaction scenarios.
{"title":"Integrating User Gaze with Verbal Instruction to Reliably Estimate Robotic Task Parameters in a Human-Robot Collaborative Environment","authors":"S. K. Paul, M. Nicolescu, M. Nicolescu","doi":"10.1145/3589572.3589580","DOIUrl":"https://doi.org/10.1145/3589572.3589580","url":null,"abstract":"As robots become more ubiquitous in our daily life, it has become very important to extract task and environmental information through more natural, meaningful, and easy-to-use interaction interfaces. Not only this helps the user to adapt to (thus trust) a robot in a collaborative environment, it can supplement the core sensory information, helping the robot make reliable decisions. This paper presents a framework that combines two natural interaction interfaces: speech and gaze to reliably infer the object of interest and the robotic task parameters. The gaze estimation module utilizes pre-defined 3D facial points and matches them to a set of extracted estimated 3D facial landmarks of the users from 2D images to infer the gaze direction. Subsequently, the verbal instructions are passed through a deep learning model to extract the information relevant to a robotic task. These extracted task parameters from verbal instructions along with the estimated gaze directions are combined to detect and/or disambiguate objects in the scene to generate the final task configurations. The proposed framework shows very promising results in integrating the relevant task parameters for the intended robotic tasks in different real-world interaction scenarios.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"76 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124519124","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Bin Yi, Jun Tang, Wenqiang Lin, Xiaohua Gao, Bing Zhou, Junjun Fang, Yulei Gao, Wenqi Li
The prediction of production process quality indicators plays an important role in product quality and production scheduling in process industries. In order to exploit the effective information contained in the massive process data, improve the prediction accuracy of production process quality indicators and apply to the changes of processing conditions, a hybrid model quality indicator migration learning prediction method based on convolutional network (CNN) and long-short-term memory (LSTM) is proposed. Massive amounts of historical process data, operational data and date data were constructed into a continuous feature matrix with a time-sliding window. The feature vectors are first extracted using CNN, and the feature vectors are constructed in a time-series sequence and used as input data for the LSTM network. Then the LSTM network is used for quality index prediction. In this process, migration learning strategy is introduced, which reduced the training time while ensuring the training accuracy. Finally, the correctness and effectiveness of the proposed method is verified by using the process data of a tobacco factory microtobacco cutting test line as a case object.
{"title":"Quality metrics prediction in process manufacturing based on CNN-LSTM transfer learning algorithm","authors":"Bin Yi, Jun Tang, Wenqiang Lin, Xiaohua Gao, Bing Zhou, Junjun Fang, Yulei Gao, Wenqi Li","doi":"10.1145/3589572.3589591","DOIUrl":"https://doi.org/10.1145/3589572.3589591","url":null,"abstract":"The prediction of production process quality indicators plays an important role in product quality and production scheduling in process industries. In order to exploit the effective information contained in the massive process data, improve the prediction accuracy of production process quality indicators and apply to the changes of processing conditions, a hybrid model quality indicator migration learning prediction method based on convolutional network (CNN) and long-short-term memory (LSTM) is proposed. Massive amounts of historical process data, operational data and date data were constructed into a continuous feature matrix with a time-sliding window. The feature vectors are first extracted using CNN, and the feature vectors are constructed in a time-series sequence and used as input data for the LSTM network. Then the LSTM network is used for quality index prediction. In this process, migration learning strategy is introduced, which reduced the training time while ensuring the training accuracy. Finally, the correctness and effectiveness of the proposed method is verified by using the process data of a tobacco factory microtobacco cutting test line as a case object.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130045792","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Bin Yi, Wenqi Li, Jun Tang, Xiaohua Gao, Bing Zhou, Xiaoli Xu, Peng Qin, Wenqiang Lin
For the complex dependencies of production data in time and space, a multi-temporal processing process quality prediction model GLSTM based on graph neural networks is proposed, which uses graph structure data to model the process relationships among production indicators, uses graph neural networks to aggregate spatial information among production indicators, and uses long and short term memory networks to model the complex dependencies of shop floor processing quality indicator sequences in time, and the experimental The results show that the model is able to achieve relative performance improvements of 5.40%, 15.04% and 0.30% compared to time series analysis methods.
{"title":"Multi-temporal process quality prediction based on graph neural network","authors":"Bin Yi, Wenqi Li, Jun Tang, Xiaohua Gao, Bing Zhou, Xiaoli Xu, Peng Qin, Wenqiang Lin","doi":"10.1145/3589572.3589599","DOIUrl":"https://doi.org/10.1145/3589572.3589599","url":null,"abstract":"For the complex dependencies of production data in time and space, a multi-temporal processing process quality prediction model GLSTM based on graph neural networks is proposed, which uses graph structure data to model the process relationships among production indicators, uses graph neural networks to aggregate spatial information among production indicators, and uses long and short term memory networks to model the complex dependencies of shop floor processing quality indicator sequences in time, and the experimental The results show that the model is able to achieve relative performance improvements of 5.40%, 15.04% and 0.30% compared to time series analysis methods.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116348438","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Effective and efficient methods for diagnosing cardiac conditions remain of significant importance and relevance in clinical cardiology. As such, advances in machine- and deep-learning technologies pave the way to high throughput approaches to automated classification of cardiac abnormalities. While there is rich literature on ECG-based classification of cardiac conditions, particularly on diagnosing Atrial Fibrillation, there is a dearth on algorithms that can effectively measure the onset and offset of atrial fibrillation events within an ECG. In this work, we show that an off-the-shelf machine learning algorithm can be trained on mathematically-computable shape signatures embedded within the local topology of ECGs to identify fibrillatory episodes in ECGs of AF patients. More precisely, we show that a topology-informed machine learning algorithm can accurately classify segments within an ECG as either resembling an atrial fibrillation event or not. Furthermore, we show that based on the model-provided classification of segments, a simple criterion may be used to determine whether the AF rhythm is paroxysmal or persistent.
{"title":"Detection of Fibrillatory Episodes in Atrial Fibrillation Rhythms via Topology-informed Machine Learning","authors":"Paul Samuel P. Ignacio","doi":"10.1145/3589572.3589576","DOIUrl":"https://doi.org/10.1145/3589572.3589576","url":null,"abstract":"Effective and efficient methods for diagnosing cardiac conditions remain of significant importance and relevance in clinical cardiology. As such, advances in machine- and deep-learning technologies pave the way to high throughput approaches to automated classification of cardiac abnormalities. While there is rich literature on ECG-based classification of cardiac conditions, particularly on diagnosing Atrial Fibrillation, there is a dearth on algorithms that can effectively measure the onset and offset of atrial fibrillation events within an ECG. In this work, we show that an off-the-shelf machine learning algorithm can be trained on mathematically-computable shape signatures embedded within the local topology of ECGs to identify fibrillatory episodes in ECGs of AF patients. More precisely, we show that a topology-informed machine learning algorithm can accurately classify segments within an ECG as either resembling an atrial fibrillation event or not. Furthermore, we show that based on the model-provided classification of segments, a simple criterion may be used to determine whether the AF rhythm is paroxysmal or persistent.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"159 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115636831","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A new type of multi-distorted invariant discrete orthogonal moments, discrete Radial-Harmonic-Fourier moments was proposed. The kernel function of the moments was composed of radial discrete orthogonal triangular function and angular Fourier complex componential factor. The relationship between discrete Radial-Harmonic-Fourier moments and Radial-Harmonic-Fourier moments was also analyzed. The experimental results indicate that the discrete Radial-Harmonic-Fourier moments have excellent image description ability and can be effectively used as invariant image features in image analysis and pattern recognition.
{"title":"Discrete Radial-Harmonic-Fourier Moments for Image Description","authors":"Kejia Wang, Ziliang Ping, Y. Sheng","doi":"10.1145/3589572.3589600","DOIUrl":"https://doi.org/10.1145/3589572.3589600","url":null,"abstract":"A new type of multi-distorted invariant discrete orthogonal moments, discrete Radial-Harmonic-Fourier moments was proposed. The kernel function of the moments was composed of radial discrete orthogonal triangular function and angular Fourier complex componential factor. The relationship between discrete Radial-Harmonic-Fourier moments and Radial-Harmonic-Fourier moments was also analyzed. The experimental results indicate that the discrete Radial-Harmonic-Fourier moments have excellent image description ability and can be effectively used as invariant image features in image analysis and pattern recognition.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128682268","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Adriel Isaiah V. Amoguis, Hernand Ang Hermida, G. J. B. Madrid, Gabriel Costes Marquez, Justin Opulencia Dy, Jose Gerardo Ortile Guerrero, J. Ilao
As levels of road traffic congestion increase relative to population density, it is becoming increasingly necessary for traffic managers to have awareness of road situations in real-time to keep up with traffic management. There are already existing techniques and applications in computer vision that traffic managers use to collect real-time telemetry, such as but not limited to vehicle counting algorithms. However, these algorithms and applications may not be lane-aware. Enabling lane awareness to these systems allows them to be more granular, which enables more in-depth telemetry such as lane usage, driver pattern recognition, and anomaly detection, among others. Lane awareness in these systems are enabled by performing lane segmentation. This study investigates two approaches to this. The first approach uses vehicle trajectories to generate aggregated trajectory maps, which are then clustered to determine trajectory lane membership and to generate representative trajectories that describes the lane. On the other hand, the second approach takes an end-to-end method and uses road lane features such as demarcation lines to segment lanes. The first approach proved to be more viable as a lane segmentation algorithm compared to the second approach as it was able to segment lanes more reliably, given enough vehicle trajectories are present.
{"title":"Road Lane Segmentation Using Vehicle Trajectory Tracking and Lane Demarcation Lines","authors":"Adriel Isaiah V. Amoguis, Hernand Ang Hermida, G. J. B. Madrid, Gabriel Costes Marquez, Justin Opulencia Dy, Jose Gerardo Ortile Guerrero, J. Ilao","doi":"10.1145/3589572.3589582","DOIUrl":"https://doi.org/10.1145/3589572.3589582","url":null,"abstract":"As levels of road traffic congestion increase relative to population density, it is becoming increasingly necessary for traffic managers to have awareness of road situations in real-time to keep up with traffic management. There are already existing techniques and applications in computer vision that traffic managers use to collect real-time telemetry, such as but not limited to vehicle counting algorithms. However, these algorithms and applications may not be lane-aware. Enabling lane awareness to these systems allows them to be more granular, which enables more in-depth telemetry such as lane usage, driver pattern recognition, and anomaly detection, among others. Lane awareness in these systems are enabled by performing lane segmentation. This study investigates two approaches to this. The first approach uses vehicle trajectories to generate aggregated trajectory maps, which are then clustered to determine trajectory lane membership and to generate representative trajectories that describes the lane. On the other hand, the second approach takes an end-to-end method and uses road lane features such as demarcation lines to segment lanes. The first approach proved to be more viable as a lane segmentation algorithm compared to the second approach as it was able to segment lanes more reliably, given enough vehicle trajectories are present.","PeriodicalId":296325,"journal":{"name":"Proceedings of the 2023 6th International Conference on Machine Vision and Applications","volume":"228 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124524930","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：