This article is concerned with the on-line counting of some harmful insects in videos in the framework of in situ video-surveillance that aims at the early detection of pest attacks in greenhouse crops. The challenges mainly concern the tiny resolution and color contrast of the insects of interest in the videos, the outdoor issues and the quasi-real time constraints. Thus, we propose a complete video-processing chain that can cope with such challenges quite satisfactorily while yielding an acceptable performance. The system has been validated off-line against many recorded videos for the whiteflie species (one potential harmful pest species).