Signal detection method and apparatus, relevant program, and storage medium storing the program

A signal detection apparatus for searching an input time-series signal for a signal portion similar to a reference time-series signal which is registered in advance and is shorter than the input time-series signal, the apparatus comprising:

• a reference feature calculating section for obtaining a reference feature time-series signal from the reference time-series signal, where the reference feature time-series signal consists of feature vectors;
• an input feature calculating section for obtaining an input feature time-series signal from the input time-series signal, where the input feature time-series signal consists of feature vectors;
• a reference feature coding section for converting the reference feature time-series signal into a reference coded time-series signal consisting of codes which indicate classifications;
• an input feature coding section for converting the input feature time-series signal into an input coded time-series signal consisting of codes which indicate classifications;
• a distortion adding section for adding a distortion to at least one of the reference time-series signal, the input time-series signal, the reference feature time-series signal, the input feature time-series signal, the reference coded time-series signal, and the input coded time-series signal;
• and
• a histogram collating section for determining a collation portion in the input coded time-series signal, generating histograms of both the reference coded time-series signal and the collation portion of the input coded time-series signal, and calculating a degree of similarity between the reference coded time-series signal and the collation portion based on the generated histograms, and
• wherein the histogram collating section determines different collation portions in the input coded time-series signal in turn, calculates the degree of similarity for each collation portion, compares the calculated degree of similarity with a predetermined target degree of similarity, and repeatedly executes the comparison for each determined collation portion, thereby determining whether the reference time-series signal is present in the relevant portion of the input time-series signal.

A program for making a computer execute a signal detecting operation of searching an input time-series signal for a signal portion similar to a reference time-series signal which is registered in advance and is shorter than the input time-series signal, the operation comprising:

• a reference feature calculating step of obtaining a reference feature time-series signal from the reference time-series signal, where the reference feature time-series signal consists of feature vectors;
• an input feature calculating step of obtaining an input feature time-series signal from the input time-series signal, where the input feature time-series signal consists of feature vectors;
• a reference feature coding step of converting the reference feature time-series signal into a reference coded time-series signal consisting of codes which indicate classifications;
• an input feature coding step of converting the input feature time-series signal into an input coded time-series signal consisting of codes which indicate classifications;
• a distortion adding step of adding a distortion to at least one of the reference time-series signal, the input time-series signal, the reference feature time-series signal, the input feature time-series signal, the reference coded time-series signal, and the input coded time-series signal;
• and
• a histogram collating step of determining a collation portion in the input coded time-series signal, generating histograms of both the reference coded time-series signal and the collation portion of the input coded time-series signal, and calculating a degree of similarity between the reference coded time-series signal and the collation portion based on the generated histograms, and
• wherein the degree of similarity is compared with a predetermined target degree of similarity, and the histogram collating step is repeatedly executed while changing the collation portion in the input coded time-series signal, thereby determining whether the reference time-series signal is present in the relevant portion of the input time-series signal.