Object recognition and localisation are important processes in computer vision and robotics. Advances in computer vision have resulted in many object recognition techniques, but most of them are computationally very intensive and require robots with powerful processing systems. For small robots, these techniques are not applicable because of the constraints of execution time. In this study, an optimised implementation of SURF based recognition technique is presented. Suitable image pre-processing techniques were developed which reduced the recognition time on small robots with limited processing resources. The recognition time was reduced from 39 seconds to 780 milliseconds. This recognition technique was adopted by a team of small robots which were given prior training to search for objects of interest in the environment. For the localisation of the robots and objects a new template, designed for passive markers based tracking, was introduced. These markers were placed on the top of each robot and they were tracked by the two ceiling mounted cameras. The information from both sources, that is ceiling mounted cameras and team of robots, was used collectively to localise the objects in the environment. The objects were localised with an error ranging from 2.8cm to 5.2cm from their actual positions in the test arena which has the dimensions of 150x163cm.

In multi-robotic systems, an approach to the coordination of multiple robots with each other is called swarm robotics. In swarm robotic systems, small size robots with limited memory and processing resources are used. Integration of vision sensors in such robots can complicate the design of the robots but at the same time, a single vision sensor can be used for multiple objectives as it provide rich surrounding information. As the vision algorithms are normally computationally demanding and robots in swarm systems has limited memory and processing capabilities, so the requirements of light weight vision algorithms also arises. In this research, the use of vision sensor information is made for achieving multiple objectives. A solution to obstacle avoidance, which is the basic requirement as robots move in a cluttered environment and also odometry which is essential for robot localization, is provided using only visual clues. The approach developed in this research is computationally less expensive and suitable for small size robots, where processing and memory constraints limit the use of computationally expensive approaches. To achieve this a library of vision algorithms is developed and customized for Blackfin processor based robotic systems.

Abstract:In reconfigurable modular robotics, when robot modules joins to form a robotic organism, they create a distributed processing environment in a unified system. This research builds on the efficient use of these distributed processing resources and presents the manner these resources can be utilised to implement distributed mosaic formation and object detection within the organism. The generation of mosaics provides surrounding awareness to the organism and helps it to localise itself with reference to the objects in the mosaics. Whereas, the detection of objects in the mosaic helps in identifying parts of the mosaic which needed processing.

No abstract

Abstract:In robotics, the object recognition approaches developed so far have proved very valuable, but their high memory and processing requirements make them suitable only for robots with high processing capability or for offline processing. When it comes to small size robots, these approaches are not effective and lightweight vision processing is adopted which causes a big drop in recognition performance. In this research, a computationally expensive, but efficient appearance-based object recognition approach is considered and tested on a small robotic platform which has limited memory and processing resources. Rather than processing the high resolution images, all the times, to perform recognition, a novel idea of switching between high and low resolutions, based on the "distance to object" is adopted. It is also shown that much of the computation time can be saved by identifying the irrelevant information in the images and avoid processing them with computationally expensive approaches. This helps to bridge the gap between the computationally expensive approaches and embedded platform with limited processing resources.