Target detection method and apparatus
AssigneeSamsung Electronics Co., Ltd.
InventorsBiao Wang, Chao Zhang, Changkyu Choi, Deheng Qian, Jae-Joon Han, Jingtao Xu, Hao Feng
Life of this patent
Estimating the term…
Abstract
A method of detecting a target includes generating an image pyramid based on an image on which a detection is to be performed; classifying candidate areas in the image pyramid using a cascade neural network; and determining a target area corresponding to a target included in the image based on the plurality of candidate areas, wherein the cascade neural network includes a plurality of neural networks, and at least one neural network among the neural networks includes parallel sub-neural networks.
Claims11
A target detection training method, comprising:
- receiving an image comprising a target;
- and
- training a cascade neural network comprising a plurality of neural networks using the image,
- wherein at least one neural network among the plurality of neural networks includes a plurality of parallel sub-neural networks classifying a sample set comprising a plurality of image areas into a plurality of positive samples and a plurality of negative samples based on a size of a target area corresponding to the target;
- training a first neural network based on the plurality of negative samples;
- and training a second neural network including the plurality of parallel sub-neural networks based on a misclassified sample, the plurality of negative samples, and the plurality of positive samples, and wherein the plurality of neural networks comprises the first neural network and the second neural network.
A method of training a cascade neural network comprising a plurality of neural networks, and at least one neural network from among the plurality of neural networks comprising parallel sub-neural networks, the method comprising:
- training the cascade neural network based on positive samples and negative samples of image areas of a target;
- repeatedly fine-tuning the trained cascade neural network based on at least one misclassified sample, the negative samples, and the positive samples;
- detecting a presence of a decrease in a detection rate of the parallel sub-neural networks included in the cascade neural network or a presence of an increase in an error rate of the parallel sub-neural networks, after each iteration of the fine tuning;
- and
- determining the cascade neural network on which fine tuning is performed immediately before the detecting of the presence of the decrease in the detection rate or the increase in the error rate to be the final trained cascade neural network.
A method of training a first neural network comprising a plurality of neural networks, and at least one neural network from among the plurality of neural networks comprising parallel sub-neural networks, the method comprising:
- classifying, by a second neural network, a sample set comprising image areas into positive samples and negative samples based on a size of a target area;
- determining a negative sample misclassified as a positive sample from among the positive samples using the first neural network;
- training the first neural network based on the misclassified sample, the negative samples, and the positive samples;
- repeatedly fine-tuning the trained first neural network based on the misclassified sample, the negative samples, and the positive samples;
- detecting a presence of a decrease in a detection rate of the parallel sub-neural networks included in the first neural network or a presence of an increase in an error rate of the parallel sub-neural networks, in response to each iteration of the fine tuning being completed;
- and
- determining the first neural network on which fine tuning is performed immediately before the detecting of the presence of the decrease in the detection rate or the increase in the error rate to be the final trained first neural network.
Dependencies are inferred from the claim text.