Objectness Scoring and Detection Proposals in Forward-Looking Sonar Images with Convolutional Neural Networks
Forward-looking sonar can capture high resolution images of underwater scenes, but their interpretation is complex. Generic object detection in such images has not been solved, specially in cases of small and unknown objects. In comparison, detection prop
- PDF / 3,811,752 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 75 Downloads / 187 Views
Abstract. Forward-looking sonar can capture high resolution images of underwater scenes, but their interpretation is complex. Generic object detection in such images has not been solved, specially in cases of small and unknown objects. In comparison, detection proposal algorithms have produced top performing object detectors in real-world color images. In this work we develop a Convolutional Neural Network that can reliably score objectness of image windows in forward-looking sonar images and by thresholding objectness, we generate detection proposals. In our dataset of marine garbage objects, we obtain 94 % recall, generating around 60 proposals per image. The biggest strength of our method is that it can generalize to previously unseen objects. We show this by detecting chain links, walls and a wrench without previous training in such objects. We strongly believe our method can be used for classindependent object detection, with many real-world applications such as chain following and mine detection. Keywords: Object detection · Detection proposals processing · Forward-looking sonar
1
·
Sonar image
Introduction
Autonomous Underwater Vehicles (AUVs) are increasingly being used for survey and exploration of underwater environments. For example, the oil and gas industry requires constant monitoring and surveying of seabed equipment, and marine researchers require similar capabilities in order to monitor ocean flora and fauna. The perception capabilities of AUVs are not comparable to land and air vehicles. Most of the perception tasks, such as object detection and recognition, are done in offline steps instead of online processing inside the vehicle. This limits the applications fields where AUVs are useful, and strongly decreases the level of autonomy that this kind of vehicles can achieve. Most of these limits on perception capabilities come directly from the underwater environment. Water absorbs and scatters light, which limits the use of c Springer International Publishing AG 2016 F. Schwenker et al. (Eds.): ANNPR 2016, LNAI 9896, pp. 209–219, 2016. DOI: 10.1007/978-3-319-46182-3 18
210
M. Valdenegro-Toro
optical cameras, specially near coasts and shores due to water turbidity and suspended material. Typical perception sensors for AUV are different kinds of Sonar, which uses acoustic waves to sense and image the environment. Acoustic waves can travel great distances on water with small attenuation, depending on frequency, but interpreting an image produced by a sonar can be challenging. One type of sonar sensor is Forward-Looking Sonar (FLS), where the sensor’s field of view looks forward, similar to an optical camera. Other kinds of sonar sensors have downward looking fields of view in order to survey the seabed. This kind of sensor is appropriate for object detection and recognition in AUVs. Object detection in sonar imagery is as challenging as other kinds of images. Methods from the Computer Vision community have been applied to this kind of images, but these kind of methods only produce class-specific obj
Data Loading...
