Information processing method, information processing device, and program

Abstract

Use a model that requires less training data. [Solution] The information processing method replaces the transformer mechanism of a model that includes an input layer for sequentially inputting time-series data, a transformer mechanism that processes the time-series data input by the input layer and performs classification processing on the time-series data, and an output layer that outputs the classification results performed by the transformer mechanism, with a randomized time warp mechanism that performs the classification processing based on a hypo subspace derived from a plurality of time elastic vectors generated based on the time-series data and a reference hypo subspace for each classification.

Claims

1. The transformer mechanism of a model including an input layer that sequentially inputs time-series data, a transformer mechanism that processes the time-series data input by the input layer and performs classification processing on the time-series data, and an output layer that outputs the classification results performed by the transformer mechanism is replaced with a randomized time warp mechanism that performs the classification processing based on a hypo subspace derived from a plurality of time elastic vectors generated based on the time-series data and a reference hypo subspace for each classification. Information processing methods.

2. Each frame of the video is used as the time-series data. The information processing method according to claim 1.

3. A skeleton extraction process is applied to the image of a person contained in each frame of the video to obtain a skeleton extraction image. The time-series data consisting of the three-dimensional coordinates of the joints of the person is obtained from the skeletal extraction image. The information processing method according to claim 1.

4. Based on the video footage taken inside the store, the classification process is performed using the randomized time warp mechanism. If the aforementioned classification process determines that the behavior is either suspicious or a request for assistance, the employee in the store will be notified. The information processing method according to claim 2 or 3.

5. Based on the video footage taken inside the store, the classification process is performed using the randomized time warp mechanism. When the classification process determines that an action is related to customer service, the type of customer service action, the date and time the action was detected, and the location within the store are stored. The information processing method according to claim 2 or 3.

6. From the aforementioned time-series data, a time-elastic vector is generated by randomly sampling multiple data points while maintaining the time order. Principal component analysis is applied to the set of time-elastic vectors consisting of multiple time-elastic vectors generated from the aforementioned time-series data to generate a hypo subspace. Multiple combinations of canonical angles with the hypo subspace are generated for each of the reference hypo subspaces that have been pre-assigned for each classification. The classification process is performed based on the similarity calculated for each of the multiple combinations of canonical angles. The information processing method according to claim 1.

7. It has a control unit, and the control unit is The transformer mechanism of a model including an input layer that sequentially inputs time-series data, a transformer mechanism that processes the time-series data input by the input layer and performs classification processing on the time-series data, and an output layer that outputs the classification results performed by the transformer mechanism is replaced with a randomized time warp mechanism that performs the classification processing based on a hypo subspace derived from a plurality of time elastic vectors generated based on the time-series data and a reference hypo subspace for each classification. Information processing device.

8. The transformer mechanism of a model including an input layer that sequentially inputs time-series data, a transformer mechanism that processes the time-series data input by the input layer and performs classification processing on the time-series data, and an output layer that outputs the classification results performed by the transformer mechanism is replaced with a randomized time warp mechanism that performs the classification processing based on a hypo subspace derived from a plurality of time elastic vectors generated based on the time-series data and a reference hypo subspace for each classification. A program that instructs a computer to perform a process.

Images