Orientation estimation is a very well known topic in many fields such as in aerospace or robotics. However, the sensors used are usually very ex- pensive, heavy and big, which make them not suitable for IoT (Internet of Things) based applications. This thesis presents a study of how different sensor fusion algorithms perform in low cost hardware and in high acceler- ation scenarios. For this purpose, an Arduino MKR1000 is used together with an accelerometer, gyroscope and magnetometer. The objective of the thesis is to choose the most suitable algorithm for the purposed practical application, which consists on attaching the device to a moving object, such as a skate board or a bike. Once the orientation is estimated, a movement recognition algorithm that was developed is able to match what trick or movement was performed. The algorithm chosen was the Madgwick one with some minor adjustments, which uses quaternions for the estimation and is very resilient when the device is under strong external accelerations.