CNN Based Monocular Depth Estimation

Abstract

In several applications, such as scene interpretation and reconstruction, precise depth measurement from images is a significant challenge. Current depth estimate techniques frequently provide fuzzy, low-resolution estimates. With the use of transfer learning, this research executes a convolutional neural network for generating a high-resolution depth map from a single RGB image. With a typical encoder-decoder architecture, when initializing the encoder, we use features extracted from high-performing pre-trained networks, as well as augmentation and training procedures that lead to more accurate outcomes. We demonstrate how, even with a very basic decoder, our approach can provide complete high-resolution depth maps. A wide number of deep learning approaches have recently been presented, and they have showed significant promise in dealing with the classical ill-posed issue. The studies are carried out using KITTI and NYU Depth v2, two widely utilized public datasets. We also examine the errors created by various models in order to expose the shortcomings of present approaches which accomplishes viable performance on KITTI besides NYU Depth v2.

An Experimental Study on Deep Learning Technique Implemented on Low Specification OpenMV Cam H7 Device

“…Swaraja et al [30] conducted experiments using 400 homogeneous and 4600 heterogeneous images in object depth detection. The results indicated that EfficientNet exhibited lower object depth detection accuracy at higher input resolutions than ResNet50.…”

“…The model will be loaded after the input image has been divided into numerous frames, and it will then be analyzed to determine an object's depth. This system uses a pre-trained model from Alhashim et al [30], which was developed using the NYU v2 dataset and the DenseNet-169 architectural encoder.…”

FPGA Implementation of CNN Based DepthEstimation Network: MiDaSNet

“…Among them, FPGAbased solutions offer many advantages, including lower power consumption while preserving higher inference speed, customization, and further enhancement possibilities via ASIC implementation. To mention a few, [3] applied two of the most common CNN-based network architectures, namely ResNet-50 and EfficientNet-B0, to solve the depth estimation problem. [9] implemented a depth-from-motion model using an optical flow algorithm on FPGA, which offers high speed and low resource consumption on hardware; however, the results are motion-dependent and show lower accuracy compared to more sophisticated NN-based methods.…”