Object detection is an important aspect for autonomous driving vehicles (ADV), which may comprise of a machine learning model that detects range classes. As the deployment ADV widens globally, variety objects to be detected increase beyond designated Continual object essentially ensure robust adaptation detect additional classes on fly. This study proposes novel continual method learns new class(es) along with cumulative memory from prior rounds avoid any catastrophic forgetting. The results...

On-road object detection is a critical component in an autonomous driving system. The safety of the vehicle can only be as good reliability on-road Thus, developing fast and robust algorithm has been primary goal many automotive industries institutes. In recent years, multi-purpose vision-based driver assistance systems have gained popularity with emergence deep neural network. A monocular camera developed to locate image plane estimate distance said real world or plane. this work, we...

An increasing number of tasks have been developed for autonomous driving and advanced driver assistance systems. However, this gives rise to the problem incorporating plural functionalities be ported into a power-constrained computing device. Therefore, objective work is alleviate complex learning procedure pixel-wise approach scene understanding. In paper, we go beyond detection semantic segmentation task as point implement it detect free space lane. Instead learning, trained single deep...

Deep learning models have revealed outstanding performance on image classification and object detection tasks. However, there is a crucial drop in when they are subject to learn new data incrementally the absence of previous training data. They suffer from catastrophic forgetting—abrupt performance. This phenomenon affects implementation artificial intelligence practical scenarios. To overcome forgetting, method has either saved memory or generated these methods computationally complex...

Audio-based automatic speech recognition as a hearing aid is susceptible to background noise and overlapping speeches. Consequently, audio-visual has been developed complement the audio input with additional visual information. However, huge improvement of neural networks in task resulted robust reliable lip reading framework that can recognize from alone. In this work, we propose model predict daily Mandarin conversation collect new Daily Conversation Lip Reading (DMCLR) dataset, consisting...

Monocular 3-D object detection is a low-cost and challenging task for autonomous vehicles robotics. Utilizing monocular image served as an auxiliary module growing concern recently. Currently, the expensive lidar stereo cameras have predominant performance on accurate detection, whereas monocular-based methods are considerably lower in performance. This gap minimized by reforming method single internal network. We exploit correlation between 2-D spaces, enabling boxes to leverage feature...

This research addresses the complex challenge of recognizing hand gestures irrespective user's body posture, a crucial issue in medical treatment for people with speech impairments and human-machine interfaces where precise gesture interpretation is vital. The aim to engineer an advanced recognition system, effective across various positions camera viewpoints. A novel flexible arrangement was employed, integrating CNN-Transformer hybrid model, leveraging strengths Convolutional Neural...

Single-board computers have gained popularity in the recent decade, largely due to immense advancements deep learning. Deep learning involves complex computational processes that are beyond capabilities of regular microcontrollers, thus necessitating use single-board computers. However, primarily designed operate efficiently low-power environments. Therefore, optimization is crucial for running algorithms effectively on In this work, we explore impact utilizing DeepStream framework run...

Abstract Thin-film copper offers excellent film texture for multilevel interconnections in integrated circuit fabrication due to its superior resistance electromigration and high electrical conductivity. To perform a chemical mechanical planarization process during semiconductor of copper, it is necessary have thorough understanding the nanomechanical properties thin-film copper. In this study, reacted passivation layers on silicon substrate wafers are investigated their under various...

Saat ini pemanfaatan transfer daya nirkabel sudah semakin marak. Hal dikarenakan memiliki kelebihan dimana penggunaannya lebih mudah tidak perlu terjadi kontak secara langsung. Pada penelitian ini, telah dibuat suatu sistem yang diperuntukkan pada stasiun pengisian baterai. Sistem terealisasi berupa prototype fungsional mampu mengisi baterai lead-acid 6 Volt dengan kapasitas 4,5 Ah. Metode digunakan adalah metode resonant inductive coupling dapat bekerja optimal frekuensi 91 kHz....

This study addresses the vulnerabilities of traditional monocular camera-based face recognition systems, emphasizing need for improved security and reliability in biometric authentication under varying environmental conditions, lighting, human poses. To counteract risk spoofing attacks using masks or static images, we introduce a multi-angle stereo camera system. system is strategically designed to capture facial imagery from multiple perspectives, thereby enhancing depth perception spatial...

Penelitian ini mengeksplorasi penerapan pemrosesan citra menggunakan Convolutional Neural Network( CNN) untuk klasifikasi luka kaki pada penderita diabetes mellitus. Diabetes dapat menyebabkan komplikasi serius, termasuk kaki, yang memerlukan identifikasi cepat pencegahan lebih lanjut. Metode melibatkan pengambilan dengan perangkat medis dan pengolahan awal( preprocessing) mempersiapkannya analisis CNN. Eksperimen menunjukkan bahwa CNN mampu mengklasifikasikan jenis tingkat akurasi tinggi....

Penerapan algoritma YOLOv7 dalam deteksi kecelakaan lalu lintas menggunakan google colab pada data training dari dataset robolow "accident 1" yang terdiri total 1522 gambar, dengan parameter batch size 1, epoch 40 dan optimizer SGD menghasilkan model tingkat precision sebesar 65.1%, recall 45.3%, mAP@.5 52.1%, mAP@.5:.95 26.4%, waktu pelatihan 2.319 hours berdasarkan hasil tersebut untuk kurang memuaskan. Analisis grafik menunjukkan bahwa mencapai lebih memuaskan, diperlukan jumlah besar...

Crowd counting plays a vital role in public safety, particularly during riot scenarios where understanding crowd dynamics is crucial for effective decision-making and risk mitigation. Accurate estimation such environments enables authorities to monitor the situation real time, allocate resources efficiently, prevent potential escalations. However, individuals scenario presents unique challenges due chaotic nature of scene, varying densities, obstructions caused by movement environmental...

Localizing objects from an image has been a vital part in autonomous driving since object localization performance directly correlate with the safety of passenger. Robust and accurate that can adapt to any environment always improved ensure safe reliable system. In this work, we propose CBNet, two-stage instance segmentation network for environment. The leverages powerful transformer as feature extractor improve performance. addition, our proposed utilizes cascade design both proposal...

Recently, deep learning has been widely employed across various domains. The Convolution Neural Network (CNN), a popular algorithm, successfully utilized in object recognition tasks, such as face recognition, vehicle and license plate recognition. However, conventional methods for may not be appropriate low-light image due to information loss the dark regions unexpected noise that can impair quality. Therefore, development of specialized techniques enhancement become major research focus...

Advance automation, digitization, and interconnectivity are driving rapid evolution in the manufacturing industry. As a result, smart has become increasingly popular, driven by integration of deep reinforcement learning (DRL) digital twins agent training. DRL proven to be particularly effective manufacturing, allowing agents respond dynamically an ever-changing environment with wide range components which leads slight variation objectives. This study proposes efficient approach tasks...

Monocular 3D object detection attempts to predict location and dimension in space using a single optical camera. The main challenge of monocular is the lack depth information infer object's distance. In this work, we propose based on CenterNet with discrete encoded orientation angle. Our proposed method able achieve 54.6% score for car class challenging Cityscapes autonomous driving dataset, outperforming prior by convincing margin.

In autonomous driving systems, the monocular 3D object detection algorithm is a crucial component. The safety of vehicles heavily depends on well-designed system. Therefore, developing robust and efficient major goal for institutes researchers. Having sense essential in robotics, as it allows system to understand its surroundings react accordingly. Compared with stereo-based Lidar-based methods, Object challenging task only utilizes 2D information generate complex features, making low-cost,...