Hemanth Korrapati

Image Sequence Partitioning for outdoor mapping

Most of the existing appearance based topological mapping algorithms produce dense topological maps in which each image stands as a node in the topological graph. Sparser maps can be built by representing groups of visually similar images as nodes of a topological graph. In this paper, we present a sparse topological mapping framework which uses Image Sequence Partitioning (ISP) techniques to group visually similar images as topological graph nodes. We present four different ISP techniques and evaluate their performance. In order to take advantage of the afore mentioned maps, we make use of Hierarchical Inverted Files (HIF) which enable efficient hierarchical loop closure. Outdoor experimental results demonstrating the sparsity, efficiency and accuracy achieved by the combination of ISP and HIF in performing loop closure are presented.

Vision-based sparse topological mapping

Most of the existing appearance-based topological mapping algorithms produce dense topological maps in which each image stands as a node in the topological graph. Sparser maps can be built by representing groups of visually similar images of a sequence as nodes of a topological graph. In this paper, we present a sparse/hierarchical topological mapping framework which uses Image Sequence Partitioning (ISP) to group visually similar images of a sequence as nodes which are then connected on the occurrence of loop closures to form a topological graph. An indexing data structure called Hierarchical Inverted File (HIF) is proposed to store the sparse maps so as to perform loop closure at the two different resolutions of the map namely the node level and image level. TFIDF weighting is combined with spatial and frequency constraints on the detected features for improved loop closure robustness. Our approach is compared with two other existing sparse mapping approaches which use ISP. Sparsity, efficiency and accuracy of the resulting maps are evaluated and compared to that of the other two techniques on publicly available outdoor omni-directional image sequences.

Hierarchical visual mapping with omnidirectional images

A topological mapping framework designed for omnidirectional images is presented. Omnidirectional images acquired by the robot are organized as places which are represented as nodes in the topological graph/map. Places are regions in the environment over which the global scene appearance of all acquired images is consistent. A hierarchical loop closure algorithm is proposed which quickly sifts through the places to retrieve the most similar places and another level of thorough similarity analysis is performed over the images belonging to the retrieved places. An Image similarity metric based on spatial shift of local image features across omnidirectional/panoramic image pairs is proposed. Newly proposed VLAD (Vector of Locally Aggregated Descriptors) descriptors have been used for loop closure at place and image levels. Accuracy and efficiency of our system are corroborated with experimental results on three publicly available datasets. It is shown that our approach achieves good loop closure recall rates even without using epi-polar geometry verification common among many other approaches.

Multi-resolution map building and loop closure with omnidirectional images

A topological mapping approach for omnidirectional images capable of answering loop closure queries at multiple resolutions is presented. The environment is mapped hierarchically using two layers. The first layer consists of individual images and the second layer represents regions of the environment composed of groups of images from the first layer. A hierarchical algorithm is formulated that exploits this map structure for an efficient and accurate loop closure without the need of geometric verification. The vital parameters of loop closure are automatically learned from training data. Performance of our loop closure algorithm is experimentally evaluated on various publicly available datasets and compared to two state of the art techniques. The results show that agreeable performance is achieved even on low quality datasets without the need for geometric verification of loop closures common among many contemporary approaches.

Vision-Based Hybrid Map Building for Mobile Robot Navigation

A hybrid mapping framework is presented in this work. The goal is to obtain better computational efficiency than pure metrical mapping techniques and better accuracy as well as usability for robot guidance and navigation compared to the topological mapping. Image sequences acquired in an environment by manually driving a robot are used to build a hierarchical map representation by using an image sequence partitioning (ISP) technique that uses local image features. The hierarchical map built can be understood as a topological map with nodes corresponding to certain regions in the environment. Each node in turn is made up of a set of images acquired in that region. These maps are further augmented with metrical information at those nodes which correspond to image subsequences acquired while the robot is turning as a part of its trajectory. Metrical information becomes invaluable during autonomous robot navigation through these places. Hence, we call the resulting maps hybrid since they primarily contain topological information and metrical information at places that are important for navigation. Experimental results obtained on a sequence acquired in an outdoor environment are provided to demonstrate our approach.

Visual Memory Update for Life-Long Mobile Robot Navigation

A central clue for implementation of visual memory based navigation strategies relies on efficient point matching between the current image and the key images of the memory. However, the visual memory may become out of date after some times because the appearance of real-world environments keeps changing. It is thus necessary to remove obsolete information and to add new data to the visual memory over time. In this paper, we propose a method based on short-term and long term memory concepts to update the visual memory of mobile robots during navigation. The results of our experiments show that using this method improves the robustness of the localization and path-following steps.

Adaptive visual memory for mobile robot navigation in dynamic environment

A central clue for implementation of visual memory based navigation strategies relies on efficient point matching between the current image and the key images of the memory. However, the visual memory may become out of date after some times because the appearance of real-world environments keeps changing. It is thus necessary to remove obsolete information and to add new data to the visual memory over time. In this paper, we propose a method based on short-term and long term memory concepts to update the visual memory of mobile robots during navigation. The results of our experiments show that using this method improves the robustness of the localization and path-following steps.