Automatic recognition of situations with moving objects in videos is crucial in various application fields. Trajectory-based methods for recognition of situations with moving objects in videos include finite state machines, hidden Markov models, dynamic time warping, etc. Methods based on dynamic programming like hidden Markov models and dynamic time warping are robust under noisy conditions but require preparation of training datasets that can be difficult in case of rare situations. Methods based on finite state machines make it possible to eliminate training stage at all but lack robustness under noisy conditions. This paper introduces a novel method of situation recognition that combines advantages of finite state machines and dynamic programming. The proposed method was benchmarked on recognition of situations with moving objects in road traffic scenarios. Experimental results confirmed efficiency of the proposed method on real videos.