Image processing to generate 3D map of the real environment

Robot can perform many actions not possible by the human such as accessibility of some places which are limited by space constraints and also some security issues. Then the robot must depend on them to complete the required task and stand alone mapping is essential for that. Robot mapping is the process by which the device that moves through the environment is going to map with the model of environment. Mobile robots are more common with evolution of technology and used in many industries as well as indoor and outdoor applications.Most of the robots don’t have the capability of learning environment automatically due to some limitations of computational power and safety regulations.Therefore predefined the path to navigate is becoming crucial in robotics. Every type of robots should have the ability of solving problems in Localization, Mapping and Path planning in order to get expected results. Therefore the mapping approach called SLAM(Simultaneous Localization and Mapping) has used when the robot doesn’t have a map already build of the environment  and not availability of external global localization sensors to localize itself such as GPS.A mobile robot needs to build a map of the environment and localize itself in the map at the same time To navigate in an unknown environment. So that it is called as the chicken Egg problem which indicate that a map is needed for localization as well as a pose estimate is need for mapping requirement. To address the both problems simultaneously SLAM has implemented for both 2D and 3D motions in robot navigation systems.

But the problem is that the existing autonomous robots having ability of path planning while generating maps avoiding obstacles may cost very high.Such robots were driven through the environment to incrementally create a map using sensors. Unreliability of the measurements due to sensor noises or some technical limitations is the mainly derived problem through the SLAM approach. So that different types of probabilistic methods are broadly used to reduce those errors. Combining sensor data with odometry is becoming as the solution for this to handle the SLAM efficiently.

According to the history of the robot navigation while generating the map there are several frameworks that have been developed time to time for autonomous robot navigation. Occupancy grid mapping technique is a probabilistic framework models the world as the discrete map.In 1993 Sabestian thrun developed neural network to build the map using probabilistic method.Forward sensor model is coming to the scene in 2001 and then the SLAM want to be developed to reduce the gap between localization and mapping.As the standard solution for SLAM, EKF-SLAM has developed and the Method called Gmapping exposed the ability of reducing variances of the robot pose estimations

Austan Eliazar has been developed a landmark free algorithm to estimate the robot pose and to store the observations acquire by sensor, called tree structure. Another approach called Tiny SLAM also developed as the very simple and efficient solution which is based on grid format map.These solutions basically extract the robot pose’s data and then estimate the pose from the features in 3D world and then store each frames together to generate the global map to navigate automatically.