Saeid Nooshabadi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Saeid Nooshabadi has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
Caoyang Jiang
Michigan Tech
7 shared publications- Daniel R. Fuhrmann
Department of Electrical and Computer Engineering, Michigan Technological University, Houghton, MI, USA
5 shared publications - Kunle T. Olutomilayo
Department of Electrical and Computer Engineering, Michigan Technological University, Houghton, MI, USA
5 shared publications - Mojtaba Bahramgiri
Department of Electrical and Computer Engineering, Michigan Technological University, Houghton, MI, USA
5 shared publications - Cesar Ramos Rodrigues
Biomedical Engineering Institute, Federal University of Santa Catarina (IEB-UFSC), Florianópolis 88040-900, Brazil
3 shared publications - Taimur Rabuske
Universidade de Lisboa
3 shared publications
Latest Publications
- 6 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-15
Hitch Angle Estimation for Trailer Backup System—An Object Detection and Tracking Approach
This paper proposes a novel technique for trailer angle detection (TAD) for use in an advanced trailer backup assistance system (TBAS) to accomplish a semi-autonomous or full-autonomous backup maneuver. TBAS incorporates a combined trailer-tow vehicle kinematic model that requires an accurate estimate of the hitch angle. The proposed computer vision (CV) and machine learning (ML) TAD model process the image frames, acquired from the rear-facing camera, to detect and track the trailer and estimate its orientation in relation to the tow vehicle. The technique is based on a deep learning object detection and computer vision tracking model to detect and track one or more identifiable objects on the trailer (the marker-lights on the front edges of the trailer in this work). The estimated positions of the detected marker-lights are used to perform the hitch angle estimation. The model detects the hitch angle within the specified limit with an acceptance rate of 98%. The model is implemented in real-time with a processing rate of more than 30 frames per second (fps).
- 6 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-15
Hitch Angle Estimation for Trailer Backup System—An Object Detection and Tracking Approach
This paper proposes a novel technique for trailer angle detection (TAD) for use in an advanced trailer backup assistance system (TBAS) to accomplish a semi-autonomous or full-autonomous backup maneuver. TBAS incorporates a combined trailer-tow vehicle kinematic model that requires an accurate estimate of the hitch angle. The proposed computer vision (CV) and machine learning (ML) TAD model process the image frames, acquired from the rear-facing camera, to detect and track the trailer and estimate its orientation in relation to the tow vehicle. The technique is based on a deep learning object detection and computer vision tracking model to detect and track one or more identifiable objects on the trailer (the marker-lights on the front edges of the trailer in this work). The estimated positions of the detected marker-lights are used to perform the hitch angle estimation. The model detects the hitch angle within the specified limit with an acceptance rate of 98%. The model is implemented in real-time with a processing rate of more than 30 frames per second (fps).
- Open Access
- 14 June 2022
- Published by MDPI AG in Journal of Imaging
- Vol. 8 (6) , 169
Mobile-PolypNet: Lightweight Colon Polyp Segmentation Network for Low-Resource Settings
Colon polyps, small clump of cells on the lining of the colon, can lead to colorectal cancer (CRC), one of the leading types of cancer globally. Hence, early detection of these polyps automatically is crucial in the prevention of CRC. The deep learning models proposed for the detection and segmentation of colorectal polyps are resource-consuming. This paper proposes a lightweight deep learning model for colorectal polyp segmentation that achieved state-of-the-art accuracy while significantly reducing the model size and complexity. The proposed deep learning autoencoder model employs a set of state-of-the-art architectural blocks and optimization objective functions to achieve the desired efficiency. The model is trained and tested on five publicly available colorectal polyp segmentation datasets (CVC-ClinicDB, CVC-ColonDB, EndoScene, Kvasir, and ETIS). We also performed ablation testing on the model to test various aspects of the autoencoder architecture. We performed the model evaluation by using most of the common image-segmentation metrics. The backbone model achieved a DICE score of 0.935 on the Kvasir dataset and 0.945 on the CVC-ClinicDB dataset, improving the accuracy by 4.12% and 5.12%, respectively, over the current state-of-the-art network, while using 88 times fewer parameters, 40 times less storage space, and being computationally 17 times more efficient. Our ablation study showed that the addition of ConvSkip in the autoencoder slightly improves the model’s performance but it was not significant (p-value = 0.815).
- 14 June 2022
- Published by MDPI AG in Journal of Imaging
- Vol. 8 (6) , 169
Open AccessMobile-PolypNet: Lightweight Colon Polyp Segmentation Network for Low-Resource Settings
Colon polyps, small clump of cells on the lining of the colon, can lead to colorectal cancer (CRC), one of the leading types of cancer globally. Hence, early detection of these polyps automatically is crucial in the prevention of CRC. The deep learning models proposed for the detection and segmentation of colorectal polyps are resource-consuming. This paper proposes a lightweight deep learning model for colorectal polyp segmentation that achieved state-of-the-art accuracy while significantly reducing the model size and complexity. The proposed deep learning autoencoder model employs a set of state-of-the-art architectural blocks and optimization objective functions to achieve the desired efficiency. The model is trained and tested on five publicly available colorectal polyp segmentation datasets (CVC-ClinicDB, CVC-ColonDB, EndoScene, Kvasir, and ETIS). We also performed ablation testing on the model to test various aspects of the autoencoder architecture. We performed the model evaluation by using most of the common image-segmentation metrics. The backbone model achieved a DICE score of 0.935 on the Kvasir dataset and 0.945 on the CVC-ClinicDB dataset, improving the accuracy by 4.12% and 5.12%, respectively, over the current state-of-the-art network, while using 88 times fewer parameters, 40 times less storage space, and being computationally 17 times more efficient. Our ablation study showed that the addition of ConvSkip in the autoencoder slightly improves the model’s performance but it was not significant (p-value = 0.815).
- Open Access
- 7 February 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 16592-16608
Efficient Density Estimation for High-Dimensional Data
Multivariate density estimation methods, typically work well in low dimensions and their extension to data analytics in high dimensions domain has proven challenging. For density estimation in high-dimensional big data domains, the non-parametric Bayesian sequential partitioning (BSP) algorithm provides an efficient way of partitioning the sample space, based on Bayesian inference. In this paper, we present a detailed analysis of BSP and provide a computationally efficient copula-transformed data structure and algorithm for use in density estimation for data analytics in high dimensions. Using the copula-transformed data structure, we implement the density estimation for marginals in both BSP and kernel density estimation (KDE) methods. The data structures and algorithm are suitably designed for most efficient rendering into parallel processing paradigms of open multi-processing (OPENMP®) and message passing interface (MPI).
- 7 February 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 16592-16608
Open AccessEfficient Density Estimation for High-Dimensional Data
Multivariate density estimation methods, typically work well in low dimensions and their extension to data analytics in high dimensions domain has proven challenging. For density estimation in high-dimensional big data domains, the non-parametric Bayesian sequential partitioning (BSP) algorithm provides an efficient way of partitioning the sample space, based on Bayesian inference. In this paper, we present a detailed analysis of BSP and provide a computationally efficient copula-transformed data structure and algorithm for use in density estimation for data analytics in high dimensions. Using the copula-transformed data structure, we implement the density estimation for marginals in both BSP and kernel density estimation (KDE) methods. The data structures and algorithm are suitably designed for most efficient rendering into parallel processing paradigms of open multi-processing (OPENMP®) and message passing interface (MPI).
- 25 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Vehicles
- Vol. 8 (2) , 1922-1933
Automotive Radar-Based Hitch Angle Tracking Technique for Trailer Backup Assistant Systems
We propose a technique for trailer angle detection (TAD) for use in advanced trailer backup assistance system (TBAS) for semi-autonomous or full-autonomous backup maneuver. TBAS incorporates a combined trailer-tow-vehicle kinematic model, which is requires an estimate of the hitch angle. The proposed radar-based TAD model, for the estimate of the hitch angle, processes reflections acquired from the mmWave radars situated at the rear side of the vehicle, to detect the trailer and track its orientation in relation to the tow-vehicle. This technique is based on the tracking of individual points in the merged radars point-cloud. Each tracked point is considered as a hitch angle estimator. Using the current and past position information of a point, the model estimates the current hitch angle. To offer an accurate and reliable estimation for the hitch angle, the model fuses the estimated hitch angle by all estimators, as well as the yaw rate of the vehicle. To track each individual radar point, the model employs the extended Kalman filter, which is robust to the noisy radar measurements. In the presence of strong and persistence reflections from the trailer, the model can track the trailer successfully and return the hitch angle with a reliability measure equal to 90%.
- 25 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Vehicles
- Vol. 8 (2) , 1922-1933
Automotive Radar-Based Hitch Angle Tracking Technique for Trailer Backup Assistant Systems
We propose a technique for trailer angle detection (TAD) for use in advanced trailer backup assistance system (TBAS) for semi-autonomous or full-autonomous backup maneuver. TBAS incorporates a combined trailer-tow-vehicle kinematic model, which is requires an estimate of the hitch angle. The proposed radar-based TAD model, for the estimate of the hitch angle, processes reflections acquired from the mmWave radars situated at the rear side of the vehicle, to detect the trailer and track its orientation in relation to the tow-vehicle. This technique is based on the tracking of individual points in the merged radars point-cloud. Each tracked point is considered as a hitch angle estimator. Using the current and past position information of a point, the model estimates the current hitch angle. To offer an accurate and reliable estimation for the hitch angle, the model fuses the estimated hitch angle by all estimators, as well as the yaw rate of the vehicle. To track each individual radar point, the model employs the extended Kalman filter, which is robust to the noisy radar measurements. In the presence of strong and persistence reflections from the trailer, the model can track the trailer successfully and return the hitch angle with a reliability measure equal to 90%.
- 1 November 2021
- Published by Elsevier in Signal Processing
- Vol. 188
Trailer angle estimation using radar point clouds
- 1 November 2021
- Published by Elsevier in Signal Processing
- Vol. 188
Trailer angle estimation using radar point clouds
- 19 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-11
Extrinsic Radar Calibration With Overlapping FoV and Hitch Ball Position Estimation
Sensor fusion, in many perception algorithms, requires detections from multiple sensors to be transformed onto a conveniently chosen coordinate system for joint processing. The position and orientation of the sensors need to be determined for the fusion procedure. Two automotive radars installed in the taillight fixtures of a truck are considered in this work and their orientation is defined with respect to the straight line connecting them. We extrinsically calibrate the radar geometry by estimating the rotation (mount angle) and translation parameters that are needed to transform the detections from the radars onto a coordinate system whose origin is at the truck’s hitch ball. This coordinate system is a convenient choice for algorithms which use radars, installed in similar locations, to track or sense the rotation of an attached trailer about the hitch ball. The calibration is performed by rotating a trailer or a platform, upon which corner reflectors (CRs) are placed, about the hitch ball in the direction of both the radars. The algorithm is based on two principles: 1) the use of common detections found in the overlapping field of view (FoV) of the radars to estimate the rotation parameters and 2) a search for the center of trailer/platform rotation to estimate the translation parameters, which define the hitch ball position relative to the radars. It is shown that the use of more radar detections tends to increase the calibration accuracy. The data collected in an experiment result in estimation errors of about 0.20° and 3 cm for the rotation and translation parameters, respectively.
- 19 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-11
Extrinsic Radar Calibration With Overlapping FoV and Hitch Ball Position Estimation
Sensor fusion, in many perception algorithms, requires detections from multiple sensors to be transformed onto a conveniently chosen coordinate system for joint processing. The position and orientation of the sensors need to be determined for the fusion procedure. Two automotive radars installed in the taillight fixtures of a truck are considered in this work and their orientation is defined with respect to the straight line connecting them. We extrinsically calibrate the radar geometry by estimating the rotation (mount angle) and translation parameters that are needed to transform the detections from the radars onto a coordinate system whose origin is at the truck’s hitch ball. This coordinate system is a convenient choice for algorithms which use radars, installed in similar locations, to track or sense the rotation of an attached trailer about the hitch ball. The calibration is performed by rotating a trailer or a platform, upon which corner reflectors (CRs) are placed, about the hitch ball in the direction of both the radars. The algorithm is based on two principles: 1) the use of common detections found in the overlapping field of view (FoV) of the radars to estimate the rotation parameters and 2) a search for the center of trailer/platform rotation to estimate the translation parameters, which define the hitch ball position relative to the radars. It is shown that the use of more radar detections tends to increase the calibration accuracy. The data collected in an experiment result in estimation errors of about 0.20° and 3 cm for the rotation and translation parameters, respectively.
- Open Access
- 19 August 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 116604-116611
Exploiting Block Structures of KKT Matrices for Efficient Solution of Convex Optimization Problems
Convex optimization solvers are widely used in the embedded systems that require sophisticated optimization algorithms including model predictive control (MPC). In this paper, we aim to reduce the online solve time of such convex optimization solvers so as to reduce the total runtime of the algorithm and make it suitable for real-time convex optimization. We exploit the property of the Karush–Kuhn–Tucker (KKT) matrix involved in the solution of the problem that only some parts of the matrix change during the solution iterations of the algorithm. Our results show that the proposed method can effectively reduce the runtime of the solvers.
- 19 August 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 116604-116611
Open AccessExploiting Block Structures of KKT Matrices for Efficient Solution of Convex Optimization Problems
Convex optimization solvers are widely used in the embedded systems that require sophisticated optimization algorithms including model predictive control (MPC). In this paper, we aim to reduce the online solve time of such convex optimization solvers so as to reduce the total runtime of the algorithm and make it suitable for real-time convex optimization. We exploit the property of the Karush–Kuhn–Tucker (KKT) matrix involved in the solution of the problem that only some parts of the matrix change during the solution iterations of the algorithm. Our results show that the proposed method can effectively reduce the runtime of the solvers.
- 5 May 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-13
Extrinsic Calibration of Radar Mount Position and Orientation With Multiple Target Configurations
Radars are part of the sensor suite installed on modern vehicles for environmental perception. The position and orientation of the radar must be known in order to transform the detections from the radar coordinate system to a vehicle coordinate system (VCS), which is a common requirement for multisensor fusion. In this work, 77-GHz automotive radar sensors are extrinsically calibrated by registering the radar detections of corner reflector targets with known locations of the targets in the VCS; the procedure estimates the position and orientation parameters needed to transform radar detections onto the VCS. Radar detections are noisy and very sparse; hence, an effort is put into achieving good calibration accuracy by taking advantage of multiple target configurations. Two multitarget methods are discussed; one models estimation error as white noise and averages multiple estimates, and another combines all observations to make the data points denser for a one-time global estimation. The methods are tested with both synthetic and experimental data. The synthetic data result shows that, with sparse data points per target configuration, the estimation errors obtained due to the global method tend to decay faster than those obtained due to the averaging method as the number of configurations increases. The experimental data obtained from just ten target configurations result in estimation errors of about 0.35° and 1 cm for the orientation and position parameters, respectively.
- 5 May 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-13
Extrinsic Calibration of Radar Mount Position and Orientation With Multiple Target Configurations
Radars are part of the sensor suite installed on modern vehicles for environmental perception. The position and orientation of the radar must be known in order to transform the detections from the radar coordinate system to a vehicle coordinate system (VCS), which is a common requirement for multisensor fusion. In this work, 77-GHz automotive radar sensors are extrinsically calibrated by registering the radar detections of corner reflector targets with known locations of the targets in the VCS; the procedure estimates the position and orientation parameters needed to transform radar detections onto the VCS. Radar detections are noisy and very sparse; hence, an effort is put into achieving good calibration accuracy by taking advantage of multiple target configurations. Two multitarget methods are discussed; one models estimation error as white noise and averages multiple estimates, and another combines all observations to make the data points denser for a one-time global estimation. The methods are tested with both synthetic and experimental data. The synthetic data result shows that, with sparse data points per target configuration, the estimation errors obtained due to the global method tend to decay faster than those obtained due to the averaging method as the number of configurations increases. The experimental data obtained from just ten target configurations result in estimation errors of about 0.35° and 1 cm for the orientation and position parameters, respectively.
- 1 October 2019
- Published by Elsevier in Journal of Parallel and Distributed Computing
- Vol. 132, 127-140
High-dimensional image descriptor matching using highly parallel KD-tree construction and approximate nearest neighbor search
- 1 October 2019
- Published by Elsevier in Journal of Parallel and Distributed Computing
- Vol. 132, 127-140
High-dimensional image descriptor matching using highly parallel KD-tree construction and approximate nearest neighbor search
- 1 September 2019
- Published by American Scientific Publishers in Journal of Low Power Electronics
- Vol. 15 (3) , 323-328
Versatile CMOS Current Conveyor for Digital VLSI Systems with Low-Voltage Power Supply
A low voltage supply CMOS current conveyor circuit for digital input signals from 0.25 V up to 1.2 V is presented. The circuit is optimized and pre-layout simulated in a 65 nm CMOS process technology. At the target design voltage of 1.2 V, the current conveyor has a propagation delay of 2.86 ns, an energy consumption of only 80.9 pJ, and energy-delay product (EDP) of 231 pJns for resistive load of 10 kΩ. Superior performance of this work is demonstrated through comparison with other similar published work at a frequency of 5 MHz. It is shown that the proposed circuit is suitable for digital signaling. The developed CMOS circuit perfoms correctly until 50 MHz and its EDP is 31 pJns at 10 kΩ.
- 1 September 2019
- Published by American Scientific Publishers in Journal of Low Power Electronics
- Vol. 15 (3) , 323-328
Versatile CMOS Current Conveyor for Digital VLSI Systems with Low-Voltage Power Supply
A low voltage supply CMOS current conveyor circuit for digital input signals from 0.25 V up to 1.2 V is presented. The circuit is optimized and pre-layout simulated in a 65 nm CMOS process technology. At the target design voltage of 1.2 V, the current conveyor has a propagation delay of 2.86 ns, an energy consumption of only 80.9 pJ, and energy-delay product (EDP) of 231 pJns for resistive load of 10 kΩ. Superior performance of this work is demonstrated through comparison with other similar published work at a frequency of 5 MHz. It is shown that the proposed circuit is suitable for digital signaling. The developed CMOS circuit perfoms correctly until 50 MHz and its EDP is 31 pJns at 10 kΩ.