Aktivite izleme için giyilebilir sistemlerin tasarımı ve sinir ağları ve veri füzyon teknikleri kullanılarak uygulamaları

Motion analysis assists experts in the evaluation of health problems including depression and immobility. Continuously monitoring is important for early detection and to prevent irreversible deformities. Video­based systems are popular due to their high accuracy. However; they are expensive and convenient for indoor use. To overcome these challenges, we presented various wearable devices; namely a smart garment, wireless smart insoles, and wireless inertial measurement units (IMUs). Designed wearable systems provide valuable information for gait analysis and activity recognition. Firstly, lower limb joint angles were predicted using smart garment and fusion techniques during activities. Video­based system was used as a reference device to evaluate the performance of applied techniques. Secondly, it was planned to determine the ideal sensor locations for activity recognition. IMU sensors were placed on various points of the human body and collected data was fed into artificial neural networks for evaluation. Thirdly, activity recognition were aimed using various deep learning neural networks, namely convolutional neural networks, long­-short term neural networks, and convolutional long­short term neural networks. Finally, activity recognition performance was evaluated for different sensor modalities including smart insoles and IMU sensors using neural networks. The results of four studies show that; i) our smart garment presents similar performance with a video based system, ii) chest and leg are the most ideal positions for activity recognition, iii) convolutional long short term neural networks outperform the other tested networks for activity recognition, and iv) smart insoles provide the same activity recognition performance with IMU sensors attached on the leg.