Movement Recognition and Direction Detection of Pedestrians

In this dissertation we developed a method that uses smartphone sensors to recognise the pedestrian movement (using the accelerometer) and direction (using the compass) accurately. The proposed approach works independent of the GPS and also improved the recognition time of pedestrians movement and direction in comparison to using the GPS.Currently, smartphones are the most common devices people carry with them along with the tablets and smartwatches. Most smartphones are equipped with different types of sensors, including an accelerometer, magnetometer, gyroscope, compass, and GPS. In this work, we use the data acquired from a smartphone’s accelerometer to identify the movement of a pedestrian and the smartphone compass to determine the direction of a pedestrian. Classification algorithms such as (J48) which is an open source java implementation of the C4.5 Decision Tree (DT) algorithm, k-nearest neighbor (kNN), Rule based (JRip) and Meta-level (bagging and boosting) are used to determine the type of movement, including slowing down to a stop, accelerating, and decelerating. We show that these classification algorithms achieve accuracies between 93.39% and 96.98% and are capable of recognising pedestrians movement within 500ms. The direction of pedestrians movement is determined by using the accelerometer and compass sensor of a smartphone. To ensure the accuracy of a pedestrian movement direction, the smartphone’s orientation must be pre-aligned to the pedestrian’s orientation. If the smartphone and the pedestrian orientation is not aligned, then the direction obtained from the compass will not represent the actual movement direction of the pedestrian. Therefore, we present an algorithm that is independent of the smartphone’s orientation that automatically aligns the smartphone orientation to the direction of movement when the pedestrian completes two steps while the smartphone is placed in a front trouser pocket. Our proposed automated algorithm reaches an accuracy of 96% and detects changes in pedestrians’ directions within 250ms. In this dissertation, we also studied the influence of the magnetic deviation on the compass while measuring the movement direction. To illustrate the influence of magnetic deviation, we designed a filter to differentiate between safe and endangered pedestrians based on their movement direction. We observed that magnetic deviations could influence the accuracy of the filter, so our algorithm compensates for these effects using the gyroscope of a smartphone.