In-vivo measurements are inherently hampered by the motion of behaving animals, particularly in case of using voltage sensitive dyes when the signal is acquired from small regions of interest to get the maximal temporal precision. Here we present an FPGA-based, clearly acousto-optical 3D online motion correction for signal extraction on high temporal resolution of 100 kHz on a volume of 500x500x400 um up to 100 Hz of motion with a residual motion under 1 um at proper signal-to-noise ratio. The correction can be combined with all scanning modes, including point-scan, squared- and ribbon-based scanning types for population and dendritic scanning, as well as raster and volume scan types (i.e. Z-Stack). Motion correction can also be combined with photostimulation to ensure precise targeting of optogenetic stimulation. Additionally, imaging and/or photostimulation can be triggered to avoid phototoxicity during long measurements when motion compensation needs to be continuous, while imaging and stimulation should be limited to the selected behaving period. The use of the closed-loop online motion compensation has been validated with recordings in the cortex of awake mice, via ~3kHz voltage imaging of Parvalbumin neurons using the Jedi2P sensor, and acquisition of 4D datasets (3 space + time).