Research On The Depth Estimation Of Camera Array Images

The world perceived by human beings’eyes is stereo, and this information processing method is a true reflection of the real world. Now we use the digital camera to get the two-dimensional plane image, and from which we can get stereo visual effects through the vision technology, such as three-dimensional reconstruction, super resolution reconstruction etc. The new camera array device will make above process more convenient and accurate.Camera array simulates multi-view, and uses light field information which was recorded from different direction in camera, and formed two-dimensional sub-images can recreate the objects’three-dimensional spatial information, so it has a great advantage on three-dimensional reconstruction and refocusing. In addition, because of sub-images record spatial information of the object in different positions, refocusing can also be realized(i.e., focusing after photograph). The key step to achieve the above functions is depth estimation of the scene.In this paper, we try to use method of traditional calibration and method of depth estimation, and calculate depth of the scene under the frame of array images. Related to the specific contents are as follows:1. We use the improved Zhang Zhengyou calibration method to calibrate this camera array. The hardware of camera array is an array template (includes 9 micro lenses). We obtain raw format files in the camera array, then use Bayer interpolation method to convert the files to RGB color images. Then we use the Zhang Zhengyou plane template calibration method, using optimized grouping method to calibrate each lens in order to get the internal and external parameters and distortion parameters of the micro lens group.2. We use the graph cut optimized algorithm to estimate the depth of the array images. Firstly, a suitable label is assigned to each pixel, then the global energy function is constructed, and next the network graph is established. Finally, the maximum flow minimum cut theorem is used to calculate the minimum cut, and then the optimal depth is obtained. Because the camera array includes a number of lenses, it can get several images. In general, it can be considered that the information of different image space is complementary, so it can get better depth estimation effect. Emulational results also show that with the increasing of the number of lenses, the effect of depth map is more accurate.