Master Thesis at Core Tech Graphics - Three ideas, you pick!

The Core Tech Graphics team are looking for two master thesis students. We have three ideas for you to pick from: 

• AI 3D Segmentation for Digital Twin 

• Producing a Panoramic Depth Map 

• Image stabilization with pan/tilt motion compensation 
   

Category 
Computer Science, Applied Mathematics, Technical Physics, Electronics Engineering or similar 
 

Scope 
Two students, 30hp 
 

AI 3D Segmentation for Digital Twin 

Security cameras can be used as input to a digital twin by creating 3D objects from the camera images using monocular depth estimation models to create meshes. However, this approach results in 3D models that only have a mesh on the part of the object that is visible from the camera.   

By achieving semantic understanding, we could generate complete 3D objects from 2D images and improve texture mapping of such objects. With semantic understanding, we could also align objects to the ground, walls, and other surfaces. To accomplish this, we need to explore various models for creating 3D objects from 2D images, potentially incorporating semantic context.   

  
Goal  

Your task is to explore alternative technologies to create 3D objects from 2D images and construct a way to update editable geometry in a digital twin. You need to investigate ways to generate the whole object from the part of the object that is visible, one might also need to investigate ways to identify existing objects in digital twin and update their surfaces or geometry accordingly.  

For the chosen methods, one can get benchmarks for performance and quality in different scenarios including typical indoors and outdoors surveillance environments.  

Producing a Panoramic Depth Map  

Knowing about depth information enables a lot of features for axis cameras. Depth information is very valuable for a camera, i.e. advanced analytics, 3D-masking, and so on. Depth information can be of many types, point-cloud, Depth Map, 3D Mesh, etc. There are also various methods to extract this information from a given scene. One could operate cameras in stereo configuration and compute the Depth information from the disparity map.   

This will produce a point cloud in the overlapping regions. In this case, we are still limited to the available field of view of the cameras.   

Goal  

You will work with multi-channel cameras with movable sensors. The movable sensors will enable computing depth using disparity using stereo config, and the overlap is swept across the scene. You will compute depth in each position using the pair of sensors while moving them in the allowed range of motion.   

The resulting depth information needs to be stitched to form a 360 Panoramic seamless Depth Map.  

Image stabilization with pan/tilt motion compensation  

In a practical use-case, security cameras in the field will be subjected to shaking due to wind and other external forces. The objective of image stabilization is to remove or reduce this undesirable motion. Successful stabilization brings many benefits, including reduced image noise, reduced bitrate, and of course a more pleasant viewing experience.  

The most direct way of detecting motion to be stabilized is to use an inertial motion sensor (gyroscope). This electronic device measures the rotational motion of the camera system. After a round of digital filtering, suitable correction vectors can be obtained. The correction may be applied on-camera using dedicated digital hardware. The result should be a frozen image with little apparent shaking.  

However, some cameras have internal motors which allow the camera system to pan and tilt around (pan/tilt/zoom or PTZ camera). In this case, it is not possible or desirable to keep the image completely frozen. Instead, it is necessary to track both the commanded motor motion and undesirable additional shaking. The objective of image stabilization then becomes to smooth out the image motion, making it follow the commanded motion more closely.  

Axis is now looking for students who are interested in researching, implementing, and testing new algorithms that solve the problem.  

Goal  

You will research the state of the art and select one or more appropriate methods to evaluate. The data sources available in the target camera system include a gyroscope, accelerometer, and motor control signals.  

The algorithm should be feasible to run directly on the camera’s on-board processors. However, for the purposes of this thesis it may be more convenient for you to prototype the algorithm in an offline lab environment on pre-recorded data series.   

OK, I am interested, what do I do now? 
You are valuable to us – how nice that you are interested in one of our proposals! There are a few things for you to keep in mind when applying. 

• Applications are accepted in both Swedish and English, and you apply via the proposal advert. 

• The announced thesis is open only to students affiliated with a Swedish University/College either directly or via an exchange program. 

• When the thesis proposal states that it includes two students working together, we would like you to apply in pairs. In these cases, send one application each but make sure to clearly state in your application who your co-applicant is. If you have any questions regarding this, please do not hesitate to contact us. 

• Please attach your CV and University/College grade summary. 

• Please state which thesis idea(s) you are interested in in your application 

Who to contact for any questions regarding the position! 

For more information, please contact Vinay Venkanagoud Patil, [email protected]