Cardiovascular diseases are the primary cause of death worldwide. At-home monitoring systems can be used to prevent them. For long-term monitoring, these systems require high detection accuracy and low power consumption. This paper introduces a new cardiac arrhythmia classification scheme that employs antidictionaries for identifying abnormal patterns in electrocardiograms. This system enables the training to be performed without data augmentation and a smaller dataset for training compared to existing literature. The proposed method is also compatible with an eventdriven implementation, that offers great potential for ultra-low power devices. The reported average detection accuracy reaches 97.97%. The system is demonstrated through simulations and implementation on an FPGA platform.