Method for stereo matching using end-to-end convolutional neural network
Disclosed is a stereo matching method for generating a disparity map from a stereo image. The stereo matching method may include obtaining a cost volume by applying a first convolutional neural network (CNN) and a second CNN to a left image and a right image, respectively, wherein the cost volume is determined based on feature maps extracted from the left image and the right image, respectively, performing normalization on the cost volume by applying a third CNN, up-sampling the normalized cost volume, and obtaining a disparity map by applying regression analysis to the up-sampled cost volume.
1. A stereo matching method for generating a disparity map from a stereo image, the method comprising:
obtaining a cost volume by applying a first convolutional neural network (CNN) and a second CNN to a first image and a second image, respectively, wherein the cost volume is determined based on feature maps extracted from the first image and the second image, respectively;
performing normalization on the cost volume by applying a third CNN;
up-sampling the normalized cost volume; and
obtaining a disparity map by applying regression analysis to the up-sampled cost volume,
wherein the first CNN includes a first Atrous CNN layer and the second CNN includes a second Atrous CNN layer,
wherein the first Atrous CNN layer and the second Atrous CNN layer include a plurality of convolutional layers having different Atrous values, and
wherein the plurality of convolutional layers having the different Atrous values are sequentially applied to a feature map output by a previous CNN layer.
2. The stereo matching method of claim 1 , wherein the plurality of convolutional layers having the different Atrous values are applied to areas having different sizes based on the Atrous values.
3. The stereo matching method of claim 1 , wherein the first CNN and the second CNN share a weight applied to each input image.
4. The stereo matching method of claim 1 , wherein the third CNN is configured as a CNN using a three-dimensional (3D) encoder-decoder.
5. The stereo matching method of claim 1 , further comprising determining depth information of an object included in the stereo image based on the disparity map.
6. A stereo matching apparatus comprising:
a processor configured to control the stereo matching apparatus;
a sensor coupled to the processor and configured to obtain a stereo image; and
a memory coupled to the processor and configured to store data, wherein the processor is configured to:
obtain a cost volume by applying a first convolutional neural network (CNN) and a second CNN to a first image and a second image, respectively, wherein the cost volume is determined based on feature maps extracted from the first image and the second image, respectively,
perform normalization on the cost volume by applying a third CNN,
up-sample the normalized cost volume, and
obtain a disparity map by applying regression analysis to the up-sampled cost volume,
wherein the first CNN includes a first Atrous CNN layer and the second CNN includes a second Atrous CNN layer,
wherein the first Atrous CNN layer and the second Atrous CNN layer include a plurality of convolutional layers having different Atrous values, and
wherein the plurality of convolutional layers having the different Atrous values are sequentially applied to a feature map output by a previous CNN layer.
7. The stereo matching apparatus of claim 6 , wherein the plurality of convolutional layers having the different Atrous values are applied to areas having different sizes based on the Atrous values.
8. The stereo matching apparatus of claim 6 , wherein the first CNN and the second CNN share a weight applied to each input image.
9. The stereo matching apparatus of claim 6 , wherein the third CNN is configured as a CNN using a three-dimensional (3D) encoder-decoder.
10. The stereo matching apparatus of claim 6 , wherein the processor determines depth information of an object included in the stereo image based on the disparity map.