The ‘You Only Look Once’ v3 (YOLOv3) method is among the most widely used deep learning-based object detection methods. It uses the k-means cluster method to estimate the initial width and height of the predicted bounding boxes. With this method, the estimated width and height are sensitive to the initial cluster centers, and the processing of large-scale datasets is time-consuming. In order to address these problems, a new cluster method for estimating the initial width and height of the predicted bounding boxes has been developed. Firstly, it randomly selects a couple of width and height values as one initial cluster center separate from the width and height of the ground truth boxes. Secondly, it constructs Markov chains based on the selected initial cluster and uses the final points of every Markov chain as the other initial centers. In the construction of Markov chains, the intersection-over-union method is used to compute the distance between the selected initial clusters and each candidate point, instead of the square root method. Finally, this method can be used to continually update the cluster center with each new set of width and height values, which are only a part of the data selected from the datasets. Our simulation results show that the new method has faster convergence speed for initializing the width and height of the predicted bounding boxes and that it can select more representative initial widths and heights of the predicted bounding boxes. Our proposed method achieves better performance than the YOLOv3 method in terms of recall, mean average precision, and F1-score.

https://www.mdpi.com/2079-9292/9/3/537/pdf

Object Detection Algorithm Based on Improved YOLOv3

Real-time smart video surveillance to manage safety: A case study of a transport mega-project

Biometric recognition systems are finding applications in more and more civilian fields because they proved to be reliable and accurate. Among the other technologies, ultrasound has the main merit of acquiring 3D images, which allows it to provide more distinctive features and gives it a high resistance to spoof attacks. This work reviews main research activities devoted to the study and development of ultrasound sensors and systems for biometric recognition purposes. Several transducer technologies and different ultrasound techniques have been experimented on for imaging biometric characteristics like fingerprints, hand vein pattern, palmprint, and hand geometry. In the paper, basic concepts on ultrasound imaging techniques and technologies are briefly recalled and, subsequently, research studies are classified according to the kind of technique used for collecting the ultrasound image. Overall, the overview demonstrates that ultrasound may compete with other technologies in the expanding market of biometrics, as the different commercial fingerprint sensors integrated in portable electronic devices like smartphones or tablets demonstrate.

Ultrasound Systems for Biometric Recognition.

This article is concerned with the numerical solution of nonlinear hyperbolic Schro¨dinger equations (NHSEs) via an efficient Haar wavelet collocation method (HWCM). The time derivative is approximated in the governing equations by the central difference scheme, while the space derivatives are replaced by finite Haar series, which transform it to full algebraic form. The experimental rate of convergence follows the theoretical statements of convergence and the conservation laws of energy and mass are also presented, which strengthens the proposed method to be convergent and conservative. The Haar wavelets based on numerical results for solitary wave shape of |φ| are discussed in detail. The proposed approach provides a fast convergent approximation to the NHSEs. The reliability and efficiency of the method are illustrated by computing the maximum error norm and the experimental rate of convergence for different problems. Comparisons are performed with various existing methods in recent literature and better performance of the proposed method is shown in various tables and figures.

Applications of Haar Wavelet-Finite Difference Hybrid Method and Its Convergence for Hyperbolic Nonlinear Schrödinger Equation with Energy and Mass Conversion

Adaptive motion estimation and sequential outline separation based moving object detection in video surveillance system

The real-time biometric systems are used to authenticate persons for wide range of security applications. In this paper, we propose implementation of fingerprint-based biometric system using Optimized 5/3 DWT architecture and Modified CORDIC-based Fast Fourier Transform (FFT). The Optimized 2D-DWT architecture is designed using Optimized 1D-DWT architectures, Memory Units and novel Controller Unit which is used to scan rows and columns of an image. The database fingerprint image is applied to the proposed Optimized 2D-DWT architecture to obtain four sub-bands of LL, LH, HL and HH. The efficient architecture of FFT is designed by using Modified CORDIC processor which generates twiddle factor angles of range $$0^{\circ }$$
 –
 $$360^{\circ }$$
 using Pre-processing Unit and Comparator Block. Further, the LL sub-band coefficients are applied to the Modified CORDIC based FFT to generate final fingerprint features. The test fingerprint features are obtained by repeating the same procedure and are used to match the database fingerprint image features using Euclidean Distance. The performance parameters of proposed architecture in terms of area utilization, speed and accuracy is compared with existing architecture to validate the obtained results.

Implementation of Fingerprint Based Biometric System Using Optimized 5/3 DWT Architecture and Modified CORDIC Based FFT

Design and Implementation of High Speed Background Subtraction Algorithm for Moving Object Detection

The real time biometric systems are used to authenticate persons for various applications. In this paper, we propose an efficient VLSI architecture for fingerprint recognition using O2D-DWT architecture and modified CORDIC-FFT. The O2D-DWT architecture is designed using only adders and shifters for high speed operation and is applied on fingerprint image to generate four sub-bands. The optimized Fast Fourier Transform (OFFT) architecture is designed by computing different twiddle factor angles using modified CORDIC processor and is applied on LL sub-band coefficients to generate final fingerprint features. The test fingerprint features are compared with features of fingerprint images in database using Euclidean Distance. It is observed that the performance of proposed architecture is better compared to existing architectures.

An efficient VLSI architecture for fingerprint recognition using O2D-DWT architecture and modified CORDIC-FFT

Discrete Wavelet Transform (DWT) is widely used in digital image and video processing due to its various advantages over other similar transform techniques. In this paper, efficient hardware architecture of Optimized Haar Wavelet Transform is proposed which is modeled using Optimized Kogge–Stone Adder/Subtractor, Optimized Controller, Buffer, Shifter and D_FF blocks. The existing Kogge–Stone Adder architecture is optimized by using Modified Carry Correction block which uses parallel architecture to reduce the computational delay. Similarly, the Controller block is optimized by using Clock Dividers and Reset Counter interdependently. To preserve the accuracy of the processed data, suitable size of intermediate bits in fractional format with the help of Q-notation is considered. The comparison results show that the proposed architecture performs better than existing ones concerning both hardware utilization and data accuracy.

Efficient FPGA architecture of optimized Haar wavelet transform for image and video processing applications

The moving object and face detection are vital requirement for real time security applications. In this paper, we propose an Adaptive Threshold based FPGA Implementation for Object and Face detection. The input Images and reference Images are preprocessed using Gaussian Filter to smoothen the high frequency components. The 2D-DWT is applied on Gaussian filter outputs and only LL bands are considered for further processing. The modified background with adaptive threshold are used to detect the object with LL band of reference image. The detected object is passed through Gaussian filter to enhance the quality of object. The matching unit is designed to recognize face from standard face database images. It is observed that the performance parameters such as percentage TSR and hardware utilizations are better compared to existing techniques.

/pdf/an-adaptive-threshold-based-fpga-implementation-for-object-2v6er35lq7.pdf

Sayantam Sarkar

Papers

Implementation of Fingerprint Based Biometric System Using Optimized 5/3 DWT Architecture and Modified CORDIC Based FFT

Design and Implementation of High Speed Background Subtraction Algorithm for Moving Object Detection

An efficient VLSI architecture for fingerprint recognition using O2D-DWT architecture and modified CORDIC-FFT

Efficient FPGA architecture of optimized Haar wavelet transform for image and video processing applications

An Adaptive Threshold based FPGA Implementation for Object and Face detection