PLISS: Detecting and Labeling Places Using Online Change-Point Detection

Abstract

Abstract-We present PLISS (Place Labeling through Image Sequence Segmentation), a novel technique for place recognition and categorization from visual cues. PLISS operates on video or image streams and works by segmenting it into pieces corresponding to distinct places in the environment. An online Bayesian change-point detection framework that detects changes to model parameters is used to segment the image stream. Unlike current place recognition methods, in addition to using previously learned place models for labeling, PLISS can also detect and learn a previously unknown place or place category in an online manner. Moreover, since both the inferred boundaries of places (change-points) and the place labels are fully probabilistic, they can indicate when the inference is uncertain. New places and categories are detected using a systematic statistical hypothesis testing framework. We present extensive experiments on a large and difficult image dataset. We validate our claims by comparing results obtained using different types of features and by comparing results from PLISS against the state of the art.

Object Templates for Visual Place Categorization

“…The result is shown in Table 2. Only using global configurations and Bayesian filtering, the result is worse than [6]. However, if we employ local object templates, our system becomes higher in terms of accuracy, leading [6] by about 1%.…”

“…1 Table 2. Comparing our system with [6]. [6]'s result is using SIFT descriptor, which is the best in his paper.…”

“…This paper also published a new dataset on VPC along with a Bag-ofWords model algorithm to solve the problem. In [6,18], the VPC problem was addressed in a novel way based on a fully probabilistic framework. They used a Bayesian change point detection algorithm to detect abrupt changes in image sequences and a Bag-of-Words model to measure place labels.…”

“…Our system dose not contain a statistical hypothesis testing part, but we can train a model for all the other categories so that we can compare our result with [6]. The result is shown in Table 2.…”

“…A state-of-the-art work on VPC [6] tackled these two problems by a generative classification system and an online Bayesian change point detection framework.…”