High-speed gated imaging methods such as time-of-flight or fluorescence lifetime imaging are key enablers for various applications such as gesture recognition, safety instrumentation, health monitoring and materials characterization. In these applications, short light pulses are used to generate and accumulate a photocurrent. Assuming linearity and timeinvariance, this system can be modeled by a convolution of the incoming photon stream with an impulse response function (IRF) followed by a time-gated integration. Knowing the IRF allows for further improved signal analysis and sensor design. The IRF can be measured by employing light sources resembling a delta distribution or broadband-tunable sinusoidal waveforms. Both these methods are difficult to realize for increasingly fast detectors. This paper discusses a deconvolution-based approach where the signal shape of the employed light source is considered and corrected for. The IRF reconstruction schemes introduced in this paper are based on a preprocessing step to invert the integration and followed by denoising and deconvolution. Different deconvolution algorithms have been investigated and compared. In particular, we investigated direct deconvolution, Wiener deconvolution and parametric estimation of a pre-defined IRFmodel using optimization. In order to evaluate the error of the different reconstruction methods in the presence of jitter and shot noise, a ground truth needs to be generated against which the deconvolution result can be compared. For this, example IRFs that resemble typical sensor behavior were defined using analytical models. Low normalized root-meansquare error (< 0.05) can be achieved with the parametric estimation. The advantages and disadvantages of each schemes are also discussed.
|