Using Growing Neural Gas Networks to Represent Visual Object Knowledge

Donatti, Guillermo S.; Würtz, Rolf P.

doi:10.1109/ictai.2009.81

Cited by 3 publications

(2 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They were successfully applied to a variety of tasks, e.g. representing object knowledge given visual input [1]. The feature vectors ξ i , ξ i+1 , .…”

Section: Growing Neural Gas For Vector Representationmentioning

confidence: 99%

Real-Time Anomaly Detection with a Growing Neural Gas

Waniek

Bremer

Conradt

2014

Artificial Neural Networks and Machine Learning – ICANN 2014

View full text Add to dashboard Cite

Abstract. We present a novel system for vision based anomaly detection in real-time environments. Our system uses an event-based vision sensor consisting of asynchronously operating pixels that is inspired by the human retina. Each pixel reports events of illumination changes, are processed in a purely event-based tracker that pursues edges of events in the input stream. The tracker estimates are used to determine whether the input events originate from anomalous or regular data. We distinguish between the two cases with a Growing Neural Gas (GNG), which is modified to suite our event-based processing pipeline. While learning of the GNG is supervised and performed offline, the detection is carried out online. We evaluate our system by inspection of fast-spinning cogwheels. Our system achieves faster than real-time speed on commodity hardware and generalizes well to other cases. The results of this paper can be applied both to technical implementations where high speed but little processing power is required, and for further investigations into event-based algorithms.

show abstract

“…They were successfully applied to a variety of tasks, e.g. representing object knowledge given visual input [1]. The feature vectors ξ i , ξ i+1 , .…”

Section: Growing Neural Gas For Vector Representationmentioning

confidence: 99%

Real-Time Anomaly Detection with a Growing Neural Gas

Waniek

Bremer

Conradt

2014

Artificial Neural Networks and Machine Learning – ICANN 2014

View full text Add to dashboard Cite

show abstract

“…Likewise, our algorithm does not produce a perfect photographic memory, but rather retains image representations, which contain meaningful information about the explored environment. The GNG algorithm was used in [5] and [6] to map 2D nodes onto an image. The generated map was then employed for visual object recognition and categorisation.…”

Section: Introductionmentioning

confidence: 99%

Building visual memories of video streams

2012

View full text Add to dashboard Cite

A real-time method that automatically creates a visual memory of a scene using the growing neural gas (GNG) algorithm is described. The memory consists of a graph where nodes encode the visual information of a video stream as a limited set of representative images. GNG nodes are automatically generated and dynamically clustered. This method could be employed by robotic platforms in exploratory and rescue missions.Introduction: We propose an algorithm to automatically create a visual memory for robots equipped with cameras surveying, monitoring or searching an area of interest. We wish to encode the visual information into a limited set of representative images in real-time and with limited computational overhead. The idea behind our approach is to provide a flexible graphical representation of visual memory to be subsequently used for the semantic description of a captured scene. Our intention is to generate a graph that could be easily shared among robots or a distributed set of computational nodes and is easy to grow in cooperation.The robot's visual memories are incrementally built using a growing neural gas (GNG) algorithm. The GNG was chosen because it provides flexibility and portability, by dynamically building a graph of a video or scene video sequence. GNG was originally introduced by Fritzke [1], as an unsupervised learning technique where no prior training is needed. GNG was proved to be superior to existing unsupervised methods, such as self-organising or Kohonen maps and K-means [2,3]. In a GNG graph, nodes can be disconnected whilst the network is evolving, creating a separation between uncorrelated memories. The number of nodes need not be fixed a priori, since they are incrementally added during execution. Insertion of new nodes ceases when a set of user defined performance criteria is met, or alternatively the maximum network size is reached. The algorithm iteratively learns to identify similarities of input data and classifies them into clusters. GNG is much more than a simple clustering algorithm, it provides a means to associate visual memories and a means to build ontologies of visual concepts.

show abstract