Object detection apparatus and object detection method

Abstract

An object detection apparatus that detects an object to be detected captured in a determination image according to a feature amount of the object to be detected preliminarily learned by the use of a learning image, the object detection apparatus including a detector 2 causing strong classifiers 21 m to operate in order of lower classification accuracy, continuing processing when the strong classifier 21 m has determined that the object to be detected is captured in the determination image, and determining that the object to be detected has not been detected without causing the strong classifier 21 m having classification accuracy higher than the aforementioned strong classifier 21 m to operate, when the strong classifier 21 m has determined that the object to be detected is not captured in the determination image, wherein the strong classifier 21 m inputs a classification result of the strong classifier 21 m having classification accuracy lower than the aforementioned strong classifier 21 m and determines whether the object to be detected is captured or not in the determination image according to the plurality of estimation values and the input classification result.

1 .- 10 . (canceled) 
     
     
         11 . An object detection apparatus that detects an object to be detected captured in a determination image according to a feature amount of the object to be detected preliminarily learned by the use of a learning image, the object detection apparatus comprising:
 a plurality of weak classifiers each calculating an estimation value indicating a possibility that the object to be detected is captured in the determination image according to the feature amount of the object to be detected;   a plurality of strong classifiers determining, with different levels of classification accuracy from one another, whether the object to be detected is captured or not in the determination image according to the plurality of feature amounts; and   a detector causing the plurality of strong classifiers to operate in order of lower classification accuracy, continuing processing when a first strong classifier among the plurality of strong classifies has determined that the object to be detected is captured in the determination image, and determining that the object to be detected has not been detected without causing a second strong classifier among the plurality of strong classifies having classification accuracy higher than the first strong classifier to operate, when the first strong classifier has determined that the object to be detected is not captured in the determination image, wherein   the first strong classifier inputs a classification score of a third strong classifier among the plurality of strong classifies having classification accuracy lower than the first strong classifier and determines whether the object to be detected is captured or not in the determination image according to the plurality of estimation values and the input classification score.   
     
     
         12 . The object detection apparatus according to  claim 11 , wherein
 the first strong classifier inputs the classification score of the third strong classifier having lower classification accuracy next to the first strong classifier.   
     
     
         13 . The object detection apparatus according to  claim 11 , wherein
 the first strong classifier determines whether the object to be detected is captured or not in the determination image according to a score of weighted voting by using a weight indicating a degree of ease of identification of each of first weak classifiers among the plurality of weak classifiers that are combined for the first strong classifier and the estimation value of each of the first weak classifiers, and according to the input classification score.   
     
     
         14 . The object detection apparatus according to  claim 11 , wherein
 the first strong classifier inputs a score of weighted voting by using weight indicating a degree of ease of identification of each of first weak classifies among the plurality of weak classifiers that are combined for the first strong classifier and the estimation value of each of the first weak classifies, as the classification score of the third strong classifier.   
     
     
         15 . The object detection apparatus according to  claim 14 , wherein
 each of the plurality of strong classifiers is provided with a weight according to the classification accuracy, and   the first strong classifier multiplies the input score of weighted voting by weight of the third strong classifier on an input side and determines whether the object to be detected is captured or not in the determination image by use of the multiplied value.   
     
     
         16 . An object detection method of an object detection apparatus that is provided with a plurality of weak classifiers each calculating an estimation value indicating a possibility that an object to be detected is captured in a determination image according to a feature value of the object to be detected and a plurality of strong classifiers determining, with different levels of classification accuracy from one another, whether the object to be detected is captured or not in the determination image according to the plurality of estimation values, the object detection method comprising:
 a carrying-out step of causing the plurality of strong classifiers, which are connected in series in order of lower classification accuracy, to carry out classification in the order of lower classification accuracy; and   a classification step of causing a first strong classifier among the plurality of strong classifies to input a classification score of a third strong classifier among the plurality of strong classifies having classification accuracy lower than the first strong classifier and to determine whether the object to be detected is captured or not in the determination image according to the plurality of estimation values and the input classification score, wherein   the carrying-out step is continued when the first strong classifier has determined that the object to be detected is captured in the determination image in the classification step, and the carrying-out step is interrupted when the first strong classifier has determined that the object to be detected is not captured in the determination image.   
     
     
         17 . The object detection method according to  claim 16 , wherein
 the first strong classifier inputs the classification score of the third strong classifier having lower classification accuracy next to the first strong classifier, in the classification step.   
     
     
         18 . The object detection method according to  claim 16 , wherein
 the first strong classifier determines whether the object to be detected is captured or not in the determination image according to a score of weighted voting by using a weight indicating a degree of ease of identification of each of first weak classifiers among the plurality of weak classifiers that are combined for the first strong classifier and the estimation value of each of the first weak classifiers, and according to the input classification score, in the classification step.   
     
     
         19 . The object detection method according to  claim 16 , wherein
 the first strong classifier inputs a score of the weighted voting by using weight indicating a degree of ease of identification of each of first weak classifies among the plurality of weak classifiers that are combined for the first strong classifier and the estimation value of each of the first weak classifies, as the classification score of the third strong classifier, in the classification step.   
     
     
         20 . The object detection method according to  claim 19 , wherein
 each of the plurality of strong classifiers is provided with a weight according to the classification accuracy, and   the first strong classifier multiplies the input score of weighted voting by weight of the third strong classifier on an input side and determines whether the object to be detected is captured or not in the determination image by use of the multiplied value, in the classification step.