MONOCULAR DEPTH

ESTIMATION
PURPOSE

 Depth information is important for autonomous systems to perceive environments and

estimate their own state. Traditional depth estimation methods, like structure from motion
and stereo vision matching, are built on feature correspondences of multiple viewpoints.
 Meanwhile, the predicted depth maps are sparse. Inferring depth information from a single
image(monocular depth estimation) is an ill-posed problem.
OBJECTIVES OF PROJECT

 To implement a design model based on the depth estimation t identify specific features of
the scenario considered as real time object identifications.
 Implementing the characteristics features such as from motion and stereo vision matching
with feature correspondences of multiple viewpoints.
SCOPE

 With the rapid development of deep neural networks, monocular depth estimation based on
deep learning has been widely studied recently and achieved promising performance in
accuracy.
 Meanwhile, dense depth maps are estimated from single images by deep neural networks
in an end-to-end manner. In order to improve the accuracy of depth estimation, different
kinds of network frameworks, loss functions and training strategies are proposed
subsequently.
 COMAPARITIVE STUDY
 Geometry-based methods:
 Recovering 3D structures from a couple of images
based on geometric constraints is popular way to
perceive depth and has been widely
investigated in recent forty years.
 Structure from motion (SfM) is a representative
method for estimating 3D structures from a series
of2D image sequences and is applied in 3D
reconstruction and SLAM successfully.
 driving industry for depth measurement[19], it can
The depth of sparse features are worked out by SfM
through feature correspondences and geo-metric
constraints between image sequences, i.e., the
accuracy of depth estimation relies heavily on the
exact feature matching and high-quality image
sequences.
 Sensor-based methods:
 Utilizing depth sensors, like RGB-D cameras and
LIDAR, is able to get the depth information of
the corresponding image directly.
 RGB-D cameras have the ability to get the pixel-
level dense depth map of RGB image directly, but
they suffer from the limited measurement range and
outdoor sunlight sensitivity [18].
 Although LIDAR is widely used in unmanned
only generate the sparse 3D map.
 Besides, the large size and power consumption of
these depth sensors (RGB-D cameras and
LIDAR) affect their applications to small
robotics, like drones.
REFERED SECTION

 IEEE JOURNAL AND CONFERENCE PAPERS

 MATHEMATICAL APPROACHES IN MACHNE LEARNNG
PROPSOED SOLUTION USING CNN AND
CRF MODEL
 Conditional random fields (CRFs) have always performed well in the field of image
semantic segmentation. Considering the continuity of depth values, researchers have begun
to apply CRFs to solve depth estimation problems and have achieved.
 Due to the particularity of monocular depth estimation, the supervised training of neural
networks is often limited by the scene. Thus, to overcome the need for ground truth data,
unsupervised training of a network is a popular research topic.
 The basic idea is to use either left and right images or inter-frame images, in combination
with epi-polar geometry and automatic encoders to solve the depth.
BLOCK DIAGRAM
….CONTD

 The design model improvises the criteria on the sematic and conditionality occurrences at
each phase of uncertainty output observed.
 Implementing a CRF model provides the conditional changes on the DEEP-NET with the
aspect of the design criteria as mention in objectives.
 Interpolation and predictions of the given model with parameters of the NET would
provide a depth estimator ensuring the different variations of the diagram model mention
below
FLOW DIAGRAM
REFERENCES

 1.Eigen, D.; Puhrsch, C.; Fergus, R. Depth Map Prediction from a Single Image Using a Multi-
ScaleDeep Network. In Proceedings of the Neural Information Processing Systems, Montreal, QC, Canada,8–
13 December 2014; pp. 2366–2374.
 2.Eigen, D.; Fergus, R. Predicting Depth, Surface Normals and Semantic Labels with a Common Multi-
ScaleConvolutional Architecture. In Proceedings of the 2015 IEEE International Conference on Computer
Vision(ICCV), Santiago, Chile, 11–18 December 2016.
 3.Liu, F.; Shen, C.; Lin, G. Deep Convolutional Neural Fields for Depth Estimation from a Single Image.In
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA,7–12
June 2015.
 4.Liu, F.; Shen, C.; Lin, G.; Reid, I. Learning Depth from Single Monocular Images Using Deep
ConvolutionalNeural Fields.IEEE Trans. Pattern Anal. Mach. Intell.2015,38, 2024–2039. [CrossRef] [PubMed]
 5.Trigueiros, P.; Ribeiro, F.; Reis, L.P. A Comparison of Machine Learning Algorithms Applied to Hand
GestureRecognition. In Proceedings of the 7th Iberian Conference on Information Systems and Technologies,
Mardin,Spain, 20–23 June 2012.