Oral hygiene maintenance through regular toothbrushing is fundamental to preventing dental caries and periodontal disease, yet adherence to recommended twice daily brushing remains suboptimal across all age groups. Objective monitoring of brushing behavior is essential for behavior modification interventions, but current self report methods are unreliable and existing wearable sensors have been limited to specialized research devices. This paper presents a novel smartwatch based toothbrushing detection system using a convolutional neural network (CNN) trained on tri axial accelerometer and gyroscope data. The system leverages the ubiquitous availability of consumer smartwatches (Apple Watch, Samsung Galaxy Watch, Wear OS devices) to classify brushing activity in real time. We collected a dataset of 2,500 minutes of labeled activity data from 50 participants (25 male, 25 female, age 22–65) performing toothbrushing alongside common confounding activities (eating, drinking, talking, hand gestures, typing). A 1D CNN architecture with 3 convolutional layers, batch normalization, and dropout achieves 94.2% overall accuracy, 93.7% sensitivity, and 94.8% specificity for toothbrushing detection on a 5 second sliding window. The model is quantized to 8 bit integer format, achieving a footprint of 1.2 MB and inference latency of 18 ms on a Samsung Galaxy Watch 5, enabling real time on device classification with minimal battery impact (3.2% daily consumption). Feature importance analysis reveals that the distinctive rhythmic hand motion pattern (dominant frequency 1.8–2.5 Hz, corresponding to 110–150 strokes per minute) is the primary discriminator from other activities. This work demonstrates that consumer smartwatches can reliably detect toothbrushing behavior without dedicated sensors or smartphone pairing, enabling passive monitoring for oral hygiene adherence research and just in time behavioral interventions.