Efficient and Adaptive Generic Object Detection Method for Indoor Navigation

Abstract

Real time object detection and avoidance is an important part of indoor and outdoor way finding and navigation for people with vision impairment in unfamiliar environments. The objects and their arrangement in both indoor and outdoor settings occasionally change. Even stationary objects, such as furniture, may move occasionally. Additionally, providing detailed geometric models for all objects in a single room can be a very difficult and computationally intensive task. When another of similar function replaces an object, completely new models may have to be developed. Hence, there is a need of highly efficient method in detecting generic objects, which will help in detecting objects in a changing environment. This paper, presents an image-based object detection algorithm based on stable features like edges and corners instead of appearance features (color, texture, etc.). Probabilistic Graphical Model (PGM) is used for feature extraction and generic geometric model is built to detect object by combining edges and corners. Furthermore, additional geometric information is employed to distinguish doors from other objects with similar size and shape (e.g. bookshelf, cabinet, etc.). Current research shows that generic object recognition is one of the most difficult and least understood tasks in computer vision.