Machine Graphics and Vision

A machine vision system for inspecting mechanical parts

2025-03-30T10:13:14+00:00

Computer vision-based inspection has become widely used in manufacturing industries for part identification, dimensional inspection, and guiding material handling systems. Defect-free production cannot be achieved with sampling inspection methods; therefore, a 100 percentage inspection approach is mandatory to meet the zero-defect goals of manufacturing industries. Achieving this is possible with advanced technologies, such as vision-based inspection systems. In this study, a vision-based inspection system is proposed for part identification, defect detection, and dimensional measurement. The system is validated using machined parts, including a Druck plate, Pressure plate, and Retainer. A part identification algorithm is developed based on a geometry search approach. The inspection algorithm classifies parts based on edge relationships, utilizing edge detection techniques to identify each part's geometric features. Surface defects are identified by analyzing the pixel intensity gradients within defective regions. The system measures part dimensions using a vision system, with results comparable to those obtained from a coordinate measuring machine.

Basketball player target tracking based on improved YOLOv5 and multi feature fusion

2025-03-24T07:50:18+00:00

Multi-target tracking has important applications in many fields including logistics and transportation, security systems and assisted driving. With the development of science and technology, multi-target tracking has also become a research hotspot in the field of sports. In this study, a multi-attention module is added to compute the target feature information of different dimensions for the leakage problem of the traditional fifth-generation single-view detection algorithm. The study adopts two-stage target detection method to speed up the detection rate, and at the same time, recursive filtering is utilized to predict the position of the athlete in the next frame of the video. The results indicated that the improved fifth generation monovision detection algorithm possessed better results for target tracking of basketball players. The running time was reduced by 21.26% compared with the traditional fifth-generation monovision detection algorithm, and the average number of images that could be processed per second was 49. The accuracy rate was as high as 98.65%, and the average homing rate was 97.21%. During the tracking process of 60 frames of basketball sports video, the computational delay was always maintained within 40 ms. It can be demonstrated that by deeply optimizing the detection algorithm, the ability to identify and locate basketball players can be significantly improved, which provides a solid data support for the analysis of players' behaviors and tactical layout in basketball games.

V3DI ensemble model for high-accuracy aerial scene classification

2025-03-30T12:02:25+00:00

Aerial images are valuable for observing land, allowing detailed examination of Earth's surface features. As remote sensing (RS) imagery becomes more abundant, there is a growing need to fully utilize these images for smarter Earth observation. Understanding large and complex RS images is crucial. Satellite image scenery categorization, which involves labeling images based on their content, has diverse applications. Deep Learning (DL), using neural networks' powerful attribute learning capabilities, has made significant strides in categorizing satellite imagery scenes. However, recent advances in DL for scenery categorization of RS images are lacking. In our study, we employed three transfer learning (TL) models - VGG16, Densenet201 (D-201), and InceptionV3(IV3) - for classifying aerial images. VGG16 achieved 94% accuracy, while D-201 and IV3 reached 97% accuracy. Combining these models into an ensemble (V3DI ensemble model) improved accuracy to an impressive 99%. This ensemble model combines individual models' classification decisions using majority voting. We demonstrate the efficiency of this approach by showing how ensemble classification accuracy surpasses that of training individual models. Additionally, we preprocess the dataset with a Gabor filter for edge enhancement and denoising to enhance the model's overall performance.

A video-based fall detection using 3D sparse convolutional neural network in elderly care services

2025-03-30T10:13:12+00:00

Falls in the elderly have become one of the major risks for the growing elderly population. Therefore, the application of automatic fall detection system for the elderly is particularly important. In recent years, a large number of deep learning methods (such as CNN) have been applied to such research. This paper proposed a sparse convolution method 3D Sparse Convolutions and the corresponding 3D Sparse Convolutional Neural Network (3D-SCNN), which can achieve faster convolution at the approximate accuracy, thereby reducing computational complexity while maintaining high accuracy in video analysis and fall detection task. Additionally, the preprocessing stage involves a dynamic key frame selection method, using the jitter buffers to adjust frame selection based on current network conditions and buffer state. To ensure feature continuity, overlapping cubes of selected frames are intentionally employed, with dynamic resizing to adapt to network dynamics and buffer states. Experiments are conducted on Multi-camera fall dataset and UR fall dataset, and the results show that its accuracy exceeds the three compared methods, and outperforms the traditional 3D-CNN methods in both accuracy and losses.