Search

Nowadays, action recognition is widely used in various aspects of life, one of which is for monitoring purposes. One of the challenges in this problem is the process of detecting and recognizing the actions of the elderly. In this research, an integrated system is proposed to detect and monitor the actions of the elderly based on Computer Vision. This system using joint skeleton coordinate data as a system input. The system can detect actions taken by the elderly based on the dataset we created. There are seven actions: standing, sitting, walking, falling, drinking, sleeping, and eating. This system uses the OpenPose library to detect skeletons from the elderly. Then for feature extraction using spatiotemporal to extract the joint position data of the elderly every 15 frames. For classification using the Support Vector Machine in this system. Then, the recognized action data will be sent by the system to the backend of the platform. The test shows that the system can detect seven actions of the elderly with an accuracy value of 87.81% on the training model used.

COVID-19 reported cases in Malaysia is increasing every day. The lab facilities for testing COVID-19 has almost reached their capacity. Traditionally, to detect the COVID-19, a swab test is used. This method will take some time and be costly. The swab test kits are very scarce, and the human resources to do this test are limited. Through modern technology, there is a probability of detecting positive COVID-19 using X-ray images with deep learning. In this paper, Convolutional Neural Network (CNN) approach is used to detect Covid-19 through digital X-ray images. The 2D convolution kernel layer consists of three layers. The first layer has a $3 \times 3$ kernel, the second part has a $5 \times 5$ kernel, and the third part has a $7 \times 7$ kernel. Then, the output will be combined into one layer. Afterwards, the concatenated layer continued with another sequential process consisting of two convolution processes, ReLU and max pooling. Next, the model is then flattened, dropout and dense. A total of 2100 positive Covid-19 and negative Covid-19 images from Github and Kaggle databases have been used in this research. Based on the experiment done, the accuracy was almost 96%.

In Semarang, Indonesia, land subsidence had exacerbated the prone area of river floods due to heavy monsoon rain and coastal floods due to sea-level rise during high tide. Monitoring land subsidence in Semarang becomes essential to prevent the coastal inundation which leads the city to be submerged by seawater. In this study, land subsidence in Semarang was mapped using time-series analysis based on Stanford Methods for Persistent Scatterer (StaMPS) on the Sentinel-1 SAR datasets from March 2017 to May 2020 in both ascending and descending tracks. Optimized Hot Spot Analysis (OHSA) was conducted on the persistent scatterer points to spatially clustered the points with a high significance level statistically. The comparison of mean vertical deformation maps between two tracks shows a good correlation between displacement patterns. The land subsidence in Semarang was mainly due to groundwater extraction for industrial use and the compaction from the young alluvium soil.

The Internet of Things (IoT) growth are rapidly in various fields such as industry 4.0, smart cities, and smart homes. Implementation of IoT for electronic assistance had been researched to increase the longevity of human life. However, not all IoT implementation as human life assistance provides action recognition monitoring on multiple elderly people, provide information such as real-time action monitoring, and real-time streaming in a mobile application. Therefore, this research intends to create a system that can receive and provide information on each elderly people who registered. The Action Recognition Backend Platform will be working as cloud computing to receive and manage input data from Edge Computing Action Recognition. This platform integrated Deep Learning, Data Analytics, Big Data Warehouse that implemented Extract, Transform, and Load (ETL) methods, communication services with MQTT, and Kafka Streaming Processor. The test result showed that the edge computing action recognition got better model accuracy performance from our last model [1], which can predict with 50,7% accuracy in 0.5 confidence threshold. Moreover, the backend platform had been successfully implemented a simple IoT paradigm and got an average delivery time of MQTT communication at 204ms, for streaming data process took an average delay of 680ms.

Robotics is a technology that grows rapidly in this 4.0 Industrial Revolution. Robotics application in the medical field refers to the high demand from the medical world for robotics technology, where on that is also needed is to handle a test tube for laboratory use. Trajectory planning on an arm manipulator with 3 Degrees of Freedom (DOF) and parallel link mechanism is a method that will be designed in this paper with the aim of moving a test tube from a test tube rack to another rack. The test will be carried out by giving 8 coordinate points as a position reference of the trajectory planning movement. As for the test results, the average error on the x-axis, y-axis, and z-axis are 1.04 cm, 0.73 cm, and 0.8 cm respectively. So, it can be concluded that the average position error between the reference position and actual position while the arm manipulator executes the trajectory is 0.85 cm.