| Popis: |
Na podlagi podatkov o gibanju stopala lahko razlikujemo med različnimi vrstami človeškega gibanja. V diplomski nalogi je predstavljen proces izdelave prototipnega merilnega sistema, predobdelave podatkov in preizkusa uspešnosti ločevanja med hojo in tekom z različnimi algoritmi strojnega učenja. Prikazana je primerjava med rezultati v primerih, ko sta na vhodu algoritma podana pospešek ali trajektorija koraka. Izdelan merilni sistem je sestavljen iz mikroračunalnika Raspberry Pi, inercialnega senzorja, 3D natisnjenega nosilca senzorja in baterije. Senzor je nameščen na peto. Pospešek stopala merimo v smereh x, y in z. Izmerjeni podatki se prenesejo na osebni računalnik, kjer jih obdelamo s skripto v programskem jeziku Python. Cilj je odprava odstopajočih meritev in šuma. Nato izvedemo dvakratno integriranje pospeška, s katerim dobimo trajektorijo stopala. Zajete podatke o hoji in teku ustrezno razdelimo na korake. Za tem v programskem okolju Weka opravimo preizkus razpoznavanja gibanja. Based on foot movement data, we can distinguish between different types of human movement. The thesis presents the process of creating a prototype measurement system, data preprocessing, and testing the performance of recognition during walking and running with various machine learning algorithms. A comparison is shown between the results, when the inputs of algorithm is acceleration or step trajectory. The manufactured measuring system consists of a Raspberry Pi microcomputer, an inertial sensor, a 3D printed sensor girder and a battery. The sensor is placed on the heel. We measure the acceleration of the foot in the x, y and z direction. The measured data is transferred to a personal computer, where it is processed with a script in the Python programming language. The goal is to eliminate outliers and noise. Then we perform a double integration of the acceleration, which gives the trajectory of the foot. The data of walking and running is appropriately divided into steps. After that, we perform a motion recognition test in the Weka software environment. |
