Figure 2 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Source publication
Human–vehicle classification is an essential component to avoiding accidents in autonomous driving. The classification technique based on the automotive radar sensor has been paid more attention by related researchers, owing to its robustness to low-light conditions and severe weather. In the paper, we propose a hybrid support vector machine–convol...
Contexts in source publication
Context 1
... flowchart of the hybrid SVM-CNN classification method is provided, as shown in Figure 2. In summary, the 2D-FFT operator is firstly performed over the raw data, and data preprocessing including CFAR and DB-SCAN is then used in the Range-Doppler images so that these physical features of underlying targets of interest are acquired for the subsequent classification. ...
Context 2
... computational complexity of the standard SVM is O(N 2 s + N s DN) [43], where N s N is the number of support vectors, D is the dimension of the input feature vector, and N is the total sample number. In the modified SVM approach, the estimate of weight vectorˆwvectorˆ vectorˆw and the relearning of the intercept b * would take the computation of O (2N 2 s + 2N s DN). With respect to the CNN in the second stage, it is known that the CNN is composed of convolutional layers, activation functions, pooling layers, and fully connected layers, and the main computational complexity lies in the multiplication operators in convolutional layers and fully connected layers. ...
Similar publications
Detecting and recognizing different kinds of urban objects is an important problem, in particular, in autonomous driving. In this context, we studied the potential of Mueller matrix polarimetry for classifying a set of relevant real-world objects: vehicles, pedestrians, traffic signs, pavements, vegetation and tree trunks. We created a database wit...
Citations
... Figures 2, 3, 4, and 5 are image processing used in this study by implementing the HSV model, morphology, and resizing on the original image. [34]. The convolution layer, pooling layer, data normalization, and fully-connected layer are used for feature extraction from the image and to simplify the image so that the machine can learn based on the image [21]. ...
... Angelov et al. [71] considered three different types of DNNs, including the CNN, the residual network, and the combination of the CNN and the RNN, for four classes of ground target recognition, and these methods were verified on experimental data. Aiming at the problem of the class-imbalance of pedestrian and vehicle recognition with limited experimental data, Wu et al. [72] introduced a hybrid SVM-CNN method; in the first stage, a modified SVM was utilized to identify vehicle targets and adjust the imbalance between pedestrians and vehicles in the limited data, while the second stage was performed using a CNN to classify the residual unclassified targets. Experimental results illustrate that the proposed architecture can promote the performance of class-imbalance recognition on the algorithm level. ...
Radar automatic target recognition (RATR) technology is fundamental but complicated system engineering that combines sensor, target, environment, and signal processing technology, etc. It plays a significant role in improving the level and capabilities of military and civilian automation. Although RATR has been successfully applied in some aspects, the complete theoretical system has not been established. At present, deep learning algorithms have received a lot of attention and have emerged as potential and feasible solutions in RATR. This paper mainly reviews related articles published between 2010 and 2022, which corresponds to the period when deep learning methods were introduced into RATR research. In this paper, the current research status of radar target characteristics is summarized, including motion, micro-motion, one-dimensional, and two-dimensional characteristics, etc. This paper reviews the progress of deep learning methods in the feature extraction and recognition of radar target characteristics in recent years, including space, air, ground, sea-surface targets, etc. Due to more and more attention and research results published in the past few years, it is hoped that this review can provide potential guidance for future research and application of deep learning in fields related to RATR.
... An automotive radar is a system for detecting obstacles around a host vehicle on the road using microwaves. Currently, automotive radars have attracted increasing attention, and have been applied for tracking the position and velocity of preceding vehicles [1][2][3][4][5][6][7][8]. As many vehicles on the road are equipped with automotive radars, radar signals emitted from the radar of other vehicles may be detected by the receiver of the host vehicle. ...
... Thus, it is essential to develop a process that can determine the presence of a signal from a true target in the received signals, and this process in known as the detection process. Thresholding is a representative and simple detection process, and various signal processing techniques, including artificial intelligence algorithms, have been utilized for the detection process [1,3,9,10]. However, because the detected measurements include false measurements, the number of detected measurements can be larger than the actual number of targets. ...
Automotive radars, which are used for preceding vehicle tracking, have attracted significant attention in recent years. However, the false measurements that occur in cluttered roadways hinders the tracking process in vehicles; thus, it is essential to develop automotive radar systems that are robust against false measurements. This study proposed a novel track formation algorithm to initialize the preceding vehicle tracking in automotive radar systems. The proposed algorithm is based on finite impulse response filtering, and exhibited significantly higher accuracy in highly cluttered environments than a conventional track formation algorithm. The excellent performance of the proposed algorithm was demonstrated using extensive simulations under real conditions.
... A deep Convolutional Autoencoder (CAE), that essentially combines the benefits of Autoencoder and CNN, is proposed in [54] to classify 12 types of aided and non-aided human activities with an accuracy of 94.2%. Researchers have also employed hybrid classifiers for micro-Doppler-based target classification, that combine both the conventional ML and DL classifiers to take advantages of both techniques [62]. ...
With the deployment of radar in versatile scenarios and a wide variety of potential targets, demand for automatic classification of various targets is increasing. The wide variety of radar signatures from physically smaller targets due to lower velocity / radar cross-section thresholds and the increased deployment of radar-based sensors do crowd the radar screen with misinterpreted targets. Micro-Doppler signatures have been widely employed by researchers for the recognition of those targets that exhibit micro-motions. This review article presents the evolution and recent advances in radar micro-Doppler based signature analysis and feature extraction. A review of the micro-Doppler-based target classification techniques along with their applications in defense and commercial sectors, has also been presented. This article provides the first review paper in the open literature that systematically covers the major steps along with the adopted practices in micro-Doppler based target recognition. Moreover, the limitations and future trends in the field are also discussed.
... Table 6 summarises the VDC systems based on lidar sensors. (Lee et al., 2017;Zhao et al., 2020) SVM-CNN Lee et al. (2017), Wu et al. (2020), Zhao et al. (2020) and Hyun and Jin (2020) Dopplerspectrum SVM-BDT (Hyun and Jin, 2020) Classification: 90% (Lee et al., 2017), 96% , 92% Zhao et al. (2020) and 96% (Hyun and Jin, 2020) Human and vehicle Private data Capobianco et al. (2017) (Klein et al., 2006). Abdullah et al. (2016) proposed a VDC system using forward scattering radar (FSR). ...
... The proposed system achieved the accuracy of 99%, 87.54%, and 82.12% for 5 m, 10 m and 20 m range (range between vehicles to the receiver) respectively. Lee et al. (2017), Wu et al. (2020), Zhao et al. (2020) and Hyun and Jin (2020) used FMCW radar for human-vehicle classification. Lee et al. (2017) and Zhao et al. (2020) applied SVM, Wu et al. (2020) applied hybrid SVM-CNN, Hyun and Jin (2020) applied hybrid SVM-BDT classification algorithm and achieved an overall classification accuracy of more than 90%, 96%, 92%, and 96% respectively. ...
... Lee et al. (2017), Wu et al. (2020), Zhao et al. (2020) and Hyun and Jin (2020) used FMCW radar for human-vehicle classification. Lee et al. (2017) and Zhao et al. (2020) applied SVM, Wu et al. (2020) applied hybrid SVM-CNN, Hyun and Jin (2020) applied hybrid SVM-BDT classification algorithm and achieved an overall classification accuracy of more than 90%, 96%, 92%, and 96% respectively. Capobianco et al. (2017) applied CNN to the signature received from FMCW radar for classification of vehicles into six classes and achieved an average classification accuracy of 96%/ Kim et al. (2020) applied you only look once (YOLO) on pre-processed radar signal for the VDC that classifies vehicle into the trailer, Car, and pedestrian and achieved an overall accuracy of 92.07%. ...
Traffic management has become a major problem in every country due to day by day increase of vehicles on road. With this enhancement, sometimes it becomes difficult to keep track of vehicles for the purpose of traffic monitoring and law nforcement. We need an intelligenttransportation system (ITS) which will help in traffic management. In this study, we presented a review of the smart transportation system that focuses on vehicle detection and classification (VDC) that is generally used in applications like congestion prediction, future road infrastructure requirement prediction, automated parking, and security enforcement. We have reviewed more than 130 papers that are published between 2010 and 2021 and found that various sensor technologies, machine learning, computer vision and deep learning techniques have been applied for the detection and classification of vehicles by many researchers. This study will provide useful directions to the researchers in selecting appropriate technologies for VDC.
... Some machine learning based research results have been previously reported [42], but with experiments conducted in a very controlled environment where the subjects were stationary. In a related work [43], a deep learning based classification method was used for differentiating a human from a vehicle on road. In this paper, a CNN classifier is used to extract salient features and discriminate the cart from human targets. ...
This article aims to propose new techniques to detect and distinguish humans from moving machines in indoor environments. Although many research efforts have been already dedicated to humans’ indoor detection, most of the work has been focused on counting people and crowd measurement for consumer business applications. Our objective is to develop a reliable approach for humans’ indoor detection and localization aiming at avoiding collisions inside a mixed Industry 4.0 manned and unmanned environment to enhance personal and equipment safety and to prevent unwanted intrusions. An original aspect of our research is that we have worked on the real-time estimation of humans’ and moving machines’ positions while addressing the problems of multipath components and noise clutter detection. A multipulse constant false alarm rate detection algorithm is also proposed for removing the misdetections due to heavy clutter components in the indoor environment. Four impulse radio ultrawideband transceivers are placed in a specific geometry and data fusion is performed to reduce the influence of multipath and noise on the detection process. A convolutional neural network (CNN) is then used to extract the patterns corresponding to a moving machine and humans and classify them accordingly. Experiments have been carried out in two different indoor environments to demonstrate the performance of the proposed algorithms.
... A deep Convolutional Autoencoder (CAE), that essentially combines the benefits of Autoencoder and CNN, is proposed in [54] to classify 12 types of aided and non-aided human activities with an accuracy of 94.2%. Researchers have also employed hybrid classifiers for micro-Doppler-based target classification, that combine both the conventional ML and DL classifiers to take advantages of both techniques [62]. ...
This is a review article discussing the progress in using micro-doppler based features for target classification.
... A deep Convolutional Autoencoder (CAE), that essentially combines the benefits of Autoencoder and CNN, is proposed in [54] to classify 12 types of aided and non-aided human activities with an accuracy of 94.2%. Researchers have also employed hybrid classifiers for micro-Doppler-based target classification, that combine both the conventional ML and DL classifiers to take advantages of both techniques [62]. ...
This is a review article discussing the progress in using micro-doppler based features for target classification.
... The functionalities of this tier in addition to the database server are depicted in Figure 8: newPro-ductEntry, openBox, sensorAlert, handleProduct, emptyPackageExit, sendToBlockchain, prediction, correlation and patterDetection. A sequence diagram in Figure 9 illustrates interactions between tier 2 functions, along with their relationships with other modules, including the occurrence of events, their blockchains recording, and the activation of desired machine learning functions (i.e., prediction [48], classification [49] and clusterization [50]). Tier 2 has the following security properties: (a) every table has a hash field; (b) there is a shadow table in the database for auditing and backup purposes; (c) a write-only account is used, and tier four applications use a specific account with read-only permission; (d) The database server uses an encrypted disk, and the shadow tables data are written to a write-once read many (WORM) media; (e) Database server has database vault function enabled. ...
As the world population increases and the need for food monoculture farms are using more and more agrochemicals, there is also an increase in the possibility of theft, misuse, environmental damage, piracy of products, and health problems. This article addresses these issues by introducing the agrochemical pervasive traceability model (APTM), which integrates machine learning, sensors, microcontrollers, gamification, and two blockchains. It contributes in two dimensions: (I) the study of the environmental, product piracy and regulatory of agrochemical control; (II) the technological dimension: application of an adequate set of sensors collecting multiple data; modeling and implementation of a system via machine learning for analyzing and predicting the behavior and use of agrochemicals; development of a scoring system via gamification for reverse use of agrochemicals; and presenting a record of transactions in a consortium of two blockchains, simultaneously. Its main advantage is to be a flexible, adaptable, and expansive model. Results indicated that the model has positive aspects, from detecting the agrochemical, its handling, and disposal, recording of transactions, and data visualization along the reverse supply chain. This study obtained a round trip time of 0.510 ms on average; data transfers between layer one and its persistence in the database were between 4 to 5 s. Thus, blockchain nodes consumed only 34 to 38% of CPU and recorded transactions between 2 to 4 s. These results point to a horizon of applicability in real situations within agricultural farms.
... An amalgam Support Vector Machine-Convolutional Neural Network (SVM-CNN) technique is utilized for class-imbalance differentiation for vehicles and pedestrians with the available partial investigational radar data. Classification between humans and vehicles is a crucial component to prevent accidents that may be fatal for autonomous driving (Wu, Gao, Lai, & Li, 2020). ...
... This technique is built on Convolutional-Neural-Network (CNN) & image up-sampling theory. A point cloud of LIDAR data up-sampling is created & converted to pixel-level depth information; depth info is connected with Red, Green, Blue (RGB) data & inputted to deep CNN (Wu, Gao, et al., 2020). ...
Research in Autonomous Driving is taking momentum due to the inherent advantages of autonomous driving systems. The main advantage being the disassociation of the driver from the vehicle reducing the human intervention. However, the Autonomous Driving System involves many subsystems which need to be integrated as a whole system. Some of the tasks include Motion Planning, Vehicle Localization, Pedestrian Detection, Traffic Sign Detection, Road-marking Detection, Automated Parking, Vehicle Cybersecurity, and System Fault Diagnosis.
This paper aims to the overview of various Machine Learning and Deep Learning Algorithms used in Autonomous Driving Architectures for different tasks like Motion Planning, Vehicle Localization, Pedestrian Detection, Traffic Sign Detection, Road-marking Detection, Automated Parking, Vehicle Cybersecurity and Fault Diagnosis. This paper surveys the technical aspects of Machine Learning and Deep Learning Algorithms used for Autonomous Driving Systems. Comparison of these algorithms is done based on the metrics like mean Intersect in over Union (mIoU), Average Precision (AP)missed detection rate, miss rate False Positives Per Image (FPPI), and average number for false frame detection.
This study contributes to picture a review of the Machine Learning and Deep Learning Algorithms used for Autonomous Driving Systems and is organized based on the different tasks of the system.
