Preliminary Analysis of Sensor Fusion Dataset for Cyclists' Gesture Recognition

Cyclists commonly employ informal signals to communicate with each other while riding on the road; however, these signals may not be universally understood by others who are not familiar with cycling culture. For this reason, cyclists' gesture recognition is critical in ensuring the safety and efficiency of urban cycling environments. It allows cyclists and other road users to interact more easily by accurately detecting gestures such as signaling turns or stopping. Furthermore, robust gesture recognition systems can improve accessibility for cyclists with disabilities, promoting inclusivity in transportation systems. In addition, these systems contribute to the advancement of intelligent transportation technologies by enabling more responsive and adaptive infrastructure tailored to cyclists' needs. In the domain of cycling gesture recognition, the availability of high-quality datasets plays a critical role in the development and evaluation of robust algorithms. However, the creation of such datasets often poses significant challenges, including the need for diverse and representative samples, labeled data, and the consideration of real-world scenarios. We present a case study involving the development of a preliminary dataset for detecting and classifying gestures in cycling scenarios. This dataset is based on data fusion techniques for accelerometer, gyroscope, and Surface ElectroMyoGraphy (sEMG) signals; it has been collected in a laboratory environment by using a low-cost development board that could also be used for the inference of efficient neural networks within a device that is capable of recognizing cyclists gesture and announce the cyclist's intention to incoming drivers and other road users. A preliminary analysis of the dataset shows the feasibility and benefits of combining sEMG and inertial measures to classify cyclist gestures.