Performance evaluation of several SLAM algorithms in a feature-based vSLAM framework

Psoriasis is a skin disease for which currently only manual methods exist for diagnosis and monitoring of the disease. The aim of the 3DHAP project is to develop a handheld stereo camera that can scan body parts in 3D and analyze skin diseases with computational techniques.

To perform a 3D visualization, visual SLAM (Simultaneous Localization And Mapping) algorithms have to be applied. Visual SLAM (or v-SLAM) uses images acquired from cameras and other image sensors to localize itself in an unknown environment and visualize a map op this environment.

V-SLAM is composed of different part i.e. image processing, estimating poses/maps(SLAM), improving poses/maps online or offline etcetera. There are different SLAM algorithms that can be used within a V-SLAM algorithm, with each their pros and cons in terms of performance. SLAM estimates sequential movement, which includes some margin of error. By solving error minimization as an optimization problem, SLAM algorithms create a more accurate map data can be generated. Computing cost is also a problem when implementing SLAM on hardware. Computation is usually performed on compact and low-energy embedded microprocessors that have limited processing power.

The aim of this project is to give a clear view of different SLAM algorithms and measure the performance of these algorithms within a VSLAM feature-based framework. The following questions will be answered in this project:
1. Which SLAM algorithms are suitable for V-SLAM using a handheld device?
2. What is the working principle of these algorithms?
3. What is the performance of these algorithms within the V-SLAM feature-based framework?