fig.1 Event Detection Model Single stream of IMU data being categorized into specific movements. The four bottom dots represent the average of the four pressure sensor quadrants.

Turning the collected surfing data into surfboards. The different attempts to accurately categorize specific surfing maneuvers based on movement data. Developing an Activity Recognition system using visual signal processing techniques. The ultimate goal of building a cluster machine learning event detection system. Methods of indexing surfboards based on shape and function.

Event Detection

One of the most important hurdles in the entire project is the process in which the movement data collected from the IMU is categorized into specific surf maneuvers. Activity recognition aims to recognize the actions and goals of one or more agents from a series of observations on the agents' actions and the environmental conditions. The previous E.D. attempts have been largely reliant on visual signal processing with a little bit of help from video data.

By splitting the surfing session into these events we can cross reference the pressure pad and build a picture of how the surfer interacts with their equipment. The act of surfing is composed of linked turns, pumps, stalls, redirections and tricks. Determining the moment when one move starts and the next begins is an important step in quantizing the sport.

The process for building this dataset will require a variety of surfers to repeat specific maneuvers with external sensors and imaging to gather snapshots corresponding to each trick. After a sufficient amount of data is collected a series of cluster models can be trained to recognize a breadth of similar events to categorize each maneuver.

Board Indexing System

Categorizing shapes for behavior. The methodology of creating an index which can be referenced to create a variety of custom surfboards. While generating shapes from the G-code was an original intention for the project it seemed that a more scalable approach would be to leverage existing shapes. By doing so we can keep the system up to date with the most current shaping innovations. Each board is divided in half and categorized by a coordinate system which corresponds to its function. ie. a wide point forward will benefit a front footed surfer while a thin tail with benefit a back footed surfer. Fitting the generative system to this index can assign you to an existing shape or to mix and match the divided shapes to create fully new custom shapes based on the surfer with the shapers expertise behind it.

fig.2 Board Index Each board in the index is categorized by a coordinate system which corresponds to event data.