The behavior of crowds is of interest in many applications, but difficult to analyze due to the complexity of the activities taking place, the number of people moving in the scene and occlusions occurring between them. This work focuses on the problem of detecting new events in crowds using an original approach that is based on properties of the data in the Fourier domain, which leads to computationally effective and fast solutions that lead to accurate results without requiring data modeling or extensive training. The PETS2009 dataset has been used for benchmarking algorithms developed for analyzing crowd behavior, such as recognizing events in them. Experiments on the PETS 2009 dataset show that the proposed approach achieves the same or better results than existing techniques in detecting new events, while requiring almost no training samples. Extensions for accurate recognition and dealing with more complex events are also proposed as areas of future research.