The nervous system encodes and processes information with the activities of neurons. The response of a single neuron is complex and depends on the interactions between its previous state, its intrinsic properties, and the stimuli it receives. Experimentally, we utilize the patch clamp technique to monitor neural membrane voltages, but the underlying stimuli, including the external current or synaptic current from other neurons, cannot be fully observed. In this paper, we used computational models as an alternative to tackle these challenges. We employed an ensemble Kalman filter to reconstruct unobserved intracellular variables and parameters only from measured membrane potentials in a CA1 pyramidal neuron model that follows Hodgkin-Huxley dynamics. We found that the tracking of intracellular neuronal voltage and current was close to their true values whether the observations are from model generated data or real experimental data. In addition, we retrieved the experimentally inaccessible dynamics of the neuron, such as the changes of sodium and potassium gating variables, which helps to understand their roles in generating action potentials. Our study provides a powerful framework for observing dynamics underlying neural activity and seeking better real-time neuronal control.