Canine-centric interactive systems: Wavelet-based IMU motion recognition in edge AI scenarios

Accurate and real-time recognition of canine motions has significant potential in entertainment computing, especially when combined with wearable devices and gamified interactions. Therefore, a lightweight canine behavior recognition method capable of operating on wearable devices is required.This paper presents a novel wavelet-based approach for lightweight machine learning (ML) on edge devices, focusing on canine motion classification via inertial measurement unit (IMU) data. Our pipeline utilizes wavelet transforms for feature extraction and applies a compact classifier to handle computational constraints common in embedded systems. Experiments demonstrate an overall accuracy of approximately 85% across a variety of dog activities, including running, jumping, and body shaking. Building on this foundation, we propose multiple gamified scenarios to encourage dog owners to engage in daily activities, such as multi-dog leaderboards, achievement badges, and real-time interaction through sound or lighting effects. We further explore camera-based features like automatic highlight capture and augmented reality overlays, enriching user experience through playful, immersive elements. The proposed approach provides a feasible solution for canine-centric entertainment systems, balancing accuracy, low power consumption, and interactive functionality.