Abstract
Automatic doors are reactive, meaning that for every person entering a fixed sensor field, there is an opening; for every person that leaves, there is a closing, and so on; Most often, automatic doors are reactive, that is, each time someone walks into a set sensor field there's a door opening, each time someone walks out there's a closing, etc. Passersby are pedestrians who only go by the entrance. As a result, a false opening is created that exchanges conditioned air and consumes mechanical life. This paper revisits actuation as pre-trigger intent forecasting: predict if an approaching pedestrian will enter before the Position conventional trigger zone and only commit once the door can open in time. One day from the ATC shopping centre to conduct the trading day. We identify a virtual aperture and exclude 1,657 outbound leavers from the trajectory dataset that would otherwise pollute the data. The class is positive, and the same number of inbound and pass-by trajectories are kept: 1,130 and 3,609, respectively. Lead-indexed logistic-regression and random-forest The random-forest AUC declines with the increase of the distance from the warning; at 1 m, it is 0.962, and at 2 m, it is 0.918, 0.812 at the mechanically constrained distance. The random-forest AUC decreases as distance from the warning increases; at 1 m, AUC is 0.962, and at 2 m, 0.812 at the mechanically constrained distance. required 3 m operating lead, and about 0.65 outside of 4m. A fold-level uncertainty analysis reveals that the random forest method performs best. The random forest method is the best one found by a fold-level uncertainty analysis. has a strong lead at 1-2 m, a lesser and less certain one at 3 m and no lead after that. Under a 95% entrant-service In the case of constraint, the 3m policy would reduce unnecessary openings from 1,324 to 614, eliminate 710 actuation cycles and save 965.5 Wh. Serving 95.5% of entrants, in the labelled day-sample. A 1m classifier is more accurate, but physically unfeasible because the door might not travel until arrival. The study thus sets up early intent prediction as a control-feasibility. A requirement, rather than just a classification improvement.