KJCKorea
Journal Central

In this paper, we propose an edge device-based automated traffic law violation detection and easy reporting system. This system automatically detects traffic law violations such as signal violations, reckless driving, not wearing a motorcycle helmet, and vehicles driving with their tail lights off at night in a road traffic environment. It collects images in real time from edge devices and analyzes the images using a high-performance deep learning model (YOLOv8) and a lane detection model (UFLDv2) to detect various traffic law violations. In particular, to reduce false detection rates, vehicles driving without taillights are detected through multi-frame continuous detection, while helmet violations on two-wheeled vehicles are processed immediately in a single frame. Additionally, the system integrates vehicle license plate recognition and an automated reporting procedure, allowing for easy reporting of traffic law violations. The proposed system demonstrated an accuracy of 83.3% and a recall of 71.4% in detecting traffic violations. Future work will focus on validation in diverse environments and model optimization to enhance commercialization potential.

Using deep neural networks such as NoPe-NeRF, it has been observed that when generating 2D synthesized images from arbitrary viewpoints—after learning 3D spatial information from 2D input images without prior camera information—the image quality in distant regions tends to degrade significantly. This paper proposes a method to address the issue of image quality degradation in distant regions of synthesized images. The degradation is presumed to result from the use of inaccurately matched point pairs with low correspondence accuracy when computing the point cloud loss function in NoPe-NeRF. To mitigate this problem, we incorporate a Probabilistic Dense Correspondence Network (PDCNet) to improve point matching accuracy between adjacent camera images. These refined correspondences are then used to compute a more reliable point cloud loss. The effectiveness of the proposed method is validated through quantitative evaluation using PSNR, SSIM, and LPIPS, as well as qualitative visual comparisons. Furthermore, to assess improvements in camera pose estimation, ATE and RPE metrics are compared before and after applying our method. The results demonstrate that the proposed approach not only enhances the visual quality of synthesized images from arbitrary viewpoints but also improves the accuracy of camera pose estimation.

In this paper, we propose an efficient dynamic workload balancing strategy which improves the performance of high-performance computing system. The key idea of this dynamic workload balancing strategy is to minimize execution time of each job and to maximize the system throughput by effectively using system resource such as CPU, memory. Also, this strategy dynamically allocates job by considering demanded memory size of executing job and workload status of each node. If an nodes, to another free nodes and reduces the waiting time and execution time of job by balancing workload of each node. Through simulation, we show that the proposed dynamic workload balancing strategy based on CPU, memory improves the performance of high-performance computing system compared to previous strategies.