Given a database of spatial trajectories reporting the movement of a set of objects in a time frame, the problem is to discover the groups of objects that stay in close proximity within a geographical area for a significant time. To deal with the problem, techniques for the discovery of collective patterns, e.g. the meeting pattern, have been proposed. Such techniques, however, impose stringent constraints on the object movement. In this paper we investigate a generalized, flexible approach that builds on the idea of expressing the collective pattern as sum of individual behaviors. We present a technique called k-Gathering for the discovery of gatherings of at least k objects, which leverages a recent method for the discovery of stop-and-move patterns. The experiments, conducted on both synthetic and real data, show that the direction is promising and that the approach can be effective also on low sampling rate trajectories.