Depth Estimation and 3D Mapping of Navigable Indoor Environments for Assistive Robots

Depth estimation is a critical feature of assistive robots to detect their surroundings and navigate safely inside confined spaces such as buildings. The estimation of depth using stereo vision, Light Detection and Ranging (LIDAR), Time-of-Flight (ToF), Synthetic Aperture Radar (SAR), and structured lighting are some of the prominent methodologies for depth measurement. The understanding of these methods and the novel approaches for depth measurement are invaluable for constructing 3D maps of indoor environments for safe and accurate indoor maneuvering and object manipulation. The present chapter articulates the working principle in each of these methods and builds foundations for implementing the depth measurement methods for assistive robots in indoor applications. In stereo vision-based depth estimation, two cameras are employed to take photos of the same scene from two separate perspectives. The technique of triangulation and the method of calculating the depth of a 3D point based on the relative distance between the object in focus and the cameras will be discussed vividly. For indoor applications, stereo vision-based depth estimation is a simple practice since it is cost-effective and is capable of offering precise depth information. In contrast, LIDAR-based depth estimation estimates the distance between a robot and the entities in its surroundings using a LIDAR sensor. LIDAR is a method that makes use of laser beams to generate the distance between the sensor and an object of interest. In this method, the time of flight for the laser is estimated with the knowledge of the speed of the laser, and thereby the distance to the point on the surface of the object is estimated. This method is mostly employed in autonomous cars, robots, and mapping applications. Due to its ability to provide accurate depth information even in reduced illumination scenarios, LIDAR-based depth estimation is a popular method for outdoor applications. LIDAR-based depth estimation provides a reliable and stable way for generating the geometry of scenes, facilitating accurate navigation and perception in a wide range of applications. In the exploration of unstructured and featureless surfaces, the Time-of-Flight (ToF) camera-based method is one of the effective methods. The time-of-flight camera generates infrared light impulses, which are then sent to the required object. It may be noted that there are no moving parts, unlike LIDAR. The time required for the light to travel to the object, reflect, and strike the camera is computed, whereby the distance of a point on the intended surface is calculated. This process is repeated for different viewpoints, whereby a 3D point cloud is generated. The benefits and the challenges in implementing ToF measurement in comparison with LIDARs are detailed. Similar to the ToF method, which employs infrared light, the Synthetic Aperture Radar (SAR) method makes use of radio waves that are transmitted onto the intended surface. The detailed principle of operation and the reasons for its large area-fast scanning at high resolutions are discussed. Additionally, the challenges in terms of accuracy and computational effort are articulated. Cloud point data are the outcome of all the methods. Assistive robots can safely navigate indoor spaces with the help of the point cloud data produced by depth estimation. The robots can detect impediments and adjust their path accordingly due to the point cloud’s 3D representation of the surrounding area. The robot can interact with entities in its environment through the utilization of the point cloud; thus tasks such as object detection and manipulation can be achieved easily. In summary, the chapter discusses the most appropriate choice of depth estimation for indoor applications for both accuracy and precision. Moreover, the chapter also discusses economical methods for generating 3D cloud points in real-time for dense and diverse illumination environments.