Virtual reality simulation has become a crucial component in the training of surgeons as it offers a safe and immersive way to develop and improve technical skills while eliminating the need for disposable resources. However, a major limitation of contemporary simulators is that they require access to an expert surgeon during training sessions in order to provide real-time performance evaluation to the trainee. Computer-based methods that provide insightful performance evaluations relative to expert trials have the potential to alleviate the requirement of the presence of an expert during all training sessions. Dynamic time warping (DTW) is a popular algorithm for gesture recognition and evaluation as it can score the similarity between signals as a measure of Euclidean distance. Traditionally, DTW uses a sliding window to identify a given segment (sDTW). However, the segment must be of the same length as the reference segment which is being sought. In gesture recognition from surgical data, this condition is hardly true. This paper proposes a new approach to identifying optimal gesture window sizes using a sliding adaptive dynamic time warping algorithm (saDTW) in which the bounds of an initially fixed window are optimized using simulated annealing in parallel with the initial sliding window. We validate the algorithm in context of gesture recognition in a surgical simulation for percutaneous kidney stone removal. Compared to the tradition sDWT algorithm, the proposed algorithm leads to a 18.45% improvement in identifying a gesture with reference to a reference segment, and a 11.88% increased accuracy in identifying the location of a given gesture embedded in a larger signal.