Driving behavior classification is an essential real-world requirement in different contexts. In traffic safety, avoiding traffic accidents by taking corrective actions against aggressive behaviors is necessary to protect drivers. Similarly, in the automotive insurance industry, distinguishing between driving behaviors is essential to adopt usage-based insurance (UBI) policies. Also, in the ridesharing industry, monitoring and evaluating driving behaviors is critical for risk assessment and service improvement. This research presents a deep learning-based solution for driving behavior classification using an optimized Stacked-LSTM model based on the signals of smartphone embedded sensors generating two different classification models: three-class and binary. Three-class classification distinguishes between normal, drowsy, and aggressive behaviors to support advanced driver-assistance systems (ADAS). Binary classification differentiates between aggressive and non-aggressive behaviors to support commercial applications, such as ridesharing services and automotive insurance services based on UBI. Our time-series classification models have been evaluated on the public UAH-DriveSet dataset. Using the proper number and type of features, the optimum factor of upsampling for the raw signals, and the optimum time-series window size, our proposed Stacked-LSTM model made a breakthrough in the F1-score when applied to the aforementioned dataset. The achieved scores are 99.49% and 99.34% for the Three-class and binary classification models, respectively. Comparisons with state-of-the-art models, our three-class classification model surpassed the highest published F1-score of 91% by 8.49% when applied to the aforementioned dataset.

In this research, the design of a new genetic algorithm (GA) is introduced to detect the locations of license plate (LP) symbols. An adaptive threshold method is applied to overcome the dynamic changes of illumination conditions when converting the image into binary. Connected component analysis technique (CCAT) is used to detect candidate objects inside the unknown image. A scale-invariant geometric relationship matrix is introduced to model the layout of symbols in any LP that simplifies system adaptability when applied in different countries. Moreover, two new crossover operators, based on sorting, are introduced, which greatly improve the convergence speed of the system. Most of the CCAT problems, such as touching or broken bodies, are minimized by modifying the GA to perform partial match until reaching an acceptable fitness value. The system is implemented using MATLAB and various image samples are experimented with to verify the distinction of the proposed system. Encouraging results with 98.4% overall accuracy are reported for two different datasets having variability in orientation, scaling, plate location, illumination, and complex background. Examples of distorted plate images are successfully detected due to the independency on the shape, color, or location of the plate.

As a result of the urgent need to immediately identify individuals through the Internet, especially given the Coronavirus (COVID-19) pandemic at present, the recognition of online handwritten signatures has quickly evolved to become an urgent and necessary matter. However, signature identification remains challenging in the pattern recognition field due to intra-class variability and inter-class similarity. Intra-class variability is a characteristic of human behavioural activities, particularly in handwriting where no two handwritten signatures of any person can exactly coincide. The inter-class similarity is also a characteristic of human movement-based activities such as handwritten signatures particularly when the number of writers is large. In this research, an optimized transfer-learning-based architecture is proposed as a highly accurate identification technique for online-signatures using ResNet18 as a feature extraction deep-learning module. The X-Y time-series signals of the signatures were initially converted into images and used in retraining the ResNet18 model to achieve relatively high accuracy. The retrained ResNet18 model was then used to extract features that possess high discriminative distances among different classes of handwritten signatures. The model’s deep layers were searched to determine the best layer that provided the most discriminative features when using a 1-nearest neighbour learning algorithm based on the cosine distance. By using an ensemble of five models trained on rotated versions of the original signatures and using only three training samples from each writer, the classification accuracy achieved 100% when applied on the genuine signatures of public datasets such as SVC 2004 TASK1 and TASK2, and a new proprietary dataset composed of 120 genuine users. When the abovementioned technique was tested on the aggregated version of the aforementioned datasets, the resultant accuracy was still above 99%. Moreover, the robustness of the technique was proven by testing the generated models trained with one dataset with the other two datasets resulting in accuracy above 99% for all combinations.

IJSER is an open access international journal or a large number of high quality and peer reviewed research publishing in all the fields of science, engineering and technology.

Even though much advancements have been achieved with regards to the recognition of handwritten characters, researchers still face difficulties with the handwritten character recognition problem, especially with the advent of new datasets like the Extended Modified National Institute of Standards and Technology dataset (EMNIST). The EMNIST dataset represents a challenge for both machine-learning and deep-learning techniques due to inter-class similarity and intra-class variability. Inter-class similarity exists because of the similarity between the shapes of certain characters in the dataset. The presence of intra-class variability is mainly due to different shapes written by different writers for the same character. In this research, we have optimized a deep residual network to achieve higher accuracy vs. the published state-of-the-art results. This approach is mainly based on the prebuilt deep residual network model ResNet18, whose architecture has been enhanced by using the optimal number of residual blocks and the optimal size of the receptive field of the first convolutional filter, the replacement of the first max-pooling filter by an average pooling filter, and the addition of a drop-out layer before the fully connected layer. A distinctive modification has been introduced by replacing the final addition layer with a depth concatenation layer, which resulted in a novel deep architecture having higher accuracy vs. the pure residual architecture. Moreover, the dataset images’ sizes have been adjusted to optimize their visibility in the network. Finally, by tuning the training hyperparameters and using rotation and shear augmentations, the proposed model outperformed the state-of-the-art models by achieving average accuracies of 95.91% and 90.90% for the Letters and Balanced dataset sections, respectively. Furthermore, the average accuracies were improved to 95.9% and 91.06% for the Letters and Balanced sections, respectively, by using a group of 5 instances of the trained models and averaging the output class probabilities.

In this paper, we embed a signature verification mechanism in a previously introduced architecture for signature recognition to detect in-distribution and out-of-distribution random forgeries. In the previous architecture, a CNN was trained on the genuine user training dataset and then used as a feature extraction module. A k-NN algorithm with cosine distance was then used to classify the unknown signatures based on the nearest cosine distance neighbor. This architecture led to higher than 99% accuracy, but without verification, because any unknown signature will converge to one of the identities of the training dataset’s users. To add a verification mechanism that differentiates between genuine and random forgeries, we use PCA to select the most discriminating features used in calculating the cosine distance between the training and testing signatures. A fixed parameter thresholding technique based on the training distances is introduced that best differentiates between the genuine and random-user signatures. Moreover, enhancement of the technique is carried out by combining the output of the Softmax layer and the last convolution layer of the ResNet18 model to get a highly discriminative representation of the handwritten signatures. Accordingly, the introduced verification mechanism resulted in very low false positive and negative rates for test signatures from inside and outside the main dataset, with an insignificant decrease in the high identification accuracy. The complete architecture has been tested on three publicly available datasets, showing superior results.