Apparatus and method for training a learning system to detect event

Apparatus for training a learning system comprising:

• an event related keyword collecting unit that collects an event related keyword inputted by a user;
• a related keyword collecting unit that collects, by transmitting the event related keyword to a word database, at least one related keyword from the word database;
• a search formula generating unit that generates at least one search formula regarding the event related keyword and related keyword;
• an image collecting unit that collect, by transmitting the search formula to an image database, at least one event related image that is related to the search formula from the image database;
• and
• a training unit trains a learning system by communicating with the learning system with a use of the event related image as training data;
• a camera image collecting unit that collects at least one camera image obtained via a camera;
• a determination criteria information receiving unit that receives at least one criterion for determination regarding the training data from the learning system by communicating with the trained learning system;
• an unsuitable image selecting unit that selects, from the camera image, at least one unsuitable image that is determined as an event nonoccurrence image based on the at least one criterion for determination;
• an unsuitable related image collecting unit that collects, by transmitting the unsuitable image to the image database, at least one unsuitable related image that is related to the unsuitable image from the image database;
• and
• a second retraining unit that retrains a learning system by communicating with the learning system with a use of the unsuitable related image as retraining data.

A method for training a learning system comprising:

• a first operation of collecting, by an event related keyword collecting unit, an event related keyword inputted by a user;
• a second operation of transmitting, by a related keyword collecting unit, the event related keyword to a word database and collecting at least one related keyword from the word database;
• a third operation of generating, by a search formula generating unit, at least one search formula that is related to the event related keyword and related keyword;
• a fourth operation of transmitting, by an event related image collecting unit, the search formula to an image database and collecting at least one event related image that is related to the search formula from the image database;
• and
• a fifth operation of training, by a training unit, a learning system by communicating with the learning system with a use of the event related image as training data;
• a sixth operation of collecting, by a test image collecting unit, at least one test image inputted by a user;
• a seventh operation of testing, by a learning system testing unit, the trained learning system based on the test image;
• a eighth operation of collecting, by a false image collecting unit, at least one false positive image and at least one false negative image front the learning system;
• a ninth operation of transmitting, by a related false image collecting unit, the false positive image and false negative image to the image database and collecting at least one false positive image that is related to the false positive image and at least one false negative image that is related to the false negative image from the image database;
• a tenth operation of labeling, by an labeling unit, the false positive related image as an event occurrence image and labeling the false negative related image as an event nonoccurrence image;
• and
• an eleventh operation of retraining, by a first retraining unit, a learning system by communicating with the a learning system with a use of the event occurrence image and event nonoccurrence image as retraining data.

A method for training a learning system comprising:

• a first operation of collecting, by an event related keyword collecting unit, an event related keyword inputted by a user;
• a second operation of transmitting, by a related keyword collecting unit, the event related keyword to a word database and collecting at least one related keyword from the word database;
• a third operation of generation, by a search formula generating unit, at least one search formula that is related to the event related keyword and related keyword;
• a fourth operation of transmitting, by an event related image collecting unit, the search formula to an image database and collecting at least one event related image that is related to the search formula from the image database;
• and
• a fifth operation of training, by a training unit, a learning system by communicating with the learning system with a use of the event related image as training data;
• a sixth operation of collecting, by a camera image collecting unit, at least one camera image obtained via a camera;
• a seventh operation of receiving, by a determination criteria information receiving unit, at least one criterion for determination regarding the training data from the learning system by communicating with the trained learning system;
• a eighteenth operation of selecting, by an unsuitable image selecting unit, when the camera image is determined based on the at least one criterion for determination, at least one unsuitable image that is determined as an event nonoccurrence image from the camera image;
• a ninth operation of transmitting, by an unsuitable related image collecting unit, the unsuitable image to the image database and collecting at least one unsuitable related image that is related to the unsuitable image from the image database;
• and
• a tenth operation of retraining, by a second retraining unit, a learning system by communicating with the learning system with a use of the unsuitable related image as retraining data.