Address Event Representation (AER) cameras capture visual inputs in the form of asynchronous discrete events and are naturally suited to be combined with Spiking Neural Networks (SNNs) for efficient neuromorphic computation. However, noise and redundancy in event streams pose challenges for efficient processing, with longer sequences often required for effective analysis. To address these issues, we propose a method called Membrane Potential Waveform-based Adaptive Slicing (MWAS), which utilizes spiking neurons to dynamically segment the event stream based on the membrane potential waveform. By adaptively recognizing peaks in the waveform, MWAS is able to capture complete segments of motion while taking advantage of the temporal sparsity of the event stream. Furthermore, to improve inference efficiency, we introduce a dynamic termination strategy that evaluates the firing sequences of Tempotron neuron slices in real time and makes fast recognition without processing the entire event stream, significantly reducing computational overhead while maintaining accuracy. Experimental results on N-MNIST, MNIST-DVS, and DVS128 Gesture datasets demonstrate that our method achieves competitive accuracy and computational efficiency while preserving performance.