Drowsiness estimating device, drowsiness estimating method, and drowsiness estimating program recording medium

A drowsiness estimating method of estimating, from a time-series signal of an eye-openness width of a subject, drowsiness of the subject, the drowsiness estimating method comprising:

• filtering the time-series signal of the eye-openness width so as to eliminate signal changes due to blinking having a time interval of 0.1 to 0.15 [sec] of the subject therefrom to produce a filtered time-series signal of the eye-openness width;
• calculating a feature from at least the filtered time-series signal of the eye-openness width;
• and
• estimating a drowsiness evaluated value from the feature to produce an estimated result,
• wherein the calculating calculates at least a variation in the filtered time-series signal of the eye-openness width within a feature calculation window width (TM) [sec] to produce the variation as a first feature,
• wherein the time-series signal of the eye-openness width is a time-series signal of fs [frames/sec],
• wherein a relationship of N=TN×fs holds, where N represents the number of frames and TN [sec] represents a filtering calculation window width,
• wherein the filtering is to filter the time-series signal of the eye-openness width per the filtering calculation window width (TN) to produce the filtered time-series signal of the eye-openness width, and
• wherein the filtering calculation window width (TN) [sec] is lain in a range between 0.1 [sec] and 1 [sec].

A non-transitory drowsiness estimating program recording medium having recorded thereon a drowsiness estimating program for causing a computer to execute processing for estimating, from a time-series signal of an eye-openness width of a subject, drowsiness of the subject, the drowsiness estimating program causing the computer to execute:

• a filtering procedure of filtering the time-series signal of the eye-openness width so as to eliminate signal changes due to blinking having a time interval of 0.1 to 0.15 [sec] of the subject therefrom to produce a filtered time-series signal of the eye-openness width;
• a feature calculation procedure of calculating a feature from at least the filtered time-series signal of the eye-openness width;
• and
• a drowsiness estimation procedure of estimating a drowsiness evaluated value from the feature to produce an estimated result,
• wherein the feature calculation procedure at least includes a first feature calculation procedure of calculating a variation in the filtered time-series signal of the eye-openness width within a feature calculation window width (TM) [sec] to produce the variation as a first feature,
• wherein the time-series signal of the eye-openness width is a time-series signal of fs [frames/sec],
• wherein a relationship of N=TN×fs holds, where N represents the number of frames and TN [sec] represents a filtering calculation window width,
• wherein the filtering procedure is of filtering the time-series signal of the eye-openness width per the filtering calculation window width (TN) to produce the filtered time-series signal of the eye-openness width, and
• wherein the filtering calculation window width (TN) [sec] is lain in a range between 0.1 [sec] and 1 [sec].