Moving object recognition system on a pedestrian crossing

The article discusses the development of a system for recognizing people at a pedestrian crossing. The recognition system includes a trained classifier and two sets of images taken from an open database containing images of city streets from outdoor cameras. Input information prepared using PictureCropper application. The classifier was tested based on the prepared sets of images. The result is shown in the figure, and after testing, the metrics were calculated, necessary to assess the effectiveness of the resulting classifier for each set. System is being developed for specialists, which will allow individual and centralized adjustment of the adjustment of pedestrian phases for each road situation, both for the local system of controllers, and in accordance with the existing algorithms of coordination systems with the expansion of such.