Hierarchal Decomposition for Unusual Fish Trajectory Detection

Hierarchal Decomposition for Unusual Fish Trajectory Detection

Cigdem Beyan (University of Edinburgh, UK) and Robert Fisher (University of Edinburgh, UK)
DOI: 10.4018/978-1-4666-9435-4.ch001
OnDemand PDF Download:


Fish behavior analysis is presented using an unusual trajectory detection method. The proposed method is based on a hierarchy which is formed using the similarity of clustered and labeled data applying hierarchal data decomposition. The fish trajectories from unconstrained underwater videos are classified as normal and unusual where normal trajectories represents common behaviors of fish and unusual trajectories represent rare behaviors. A new trajectory is classified using the constructed hierarchy where different heuristics are applicable. The main contribution of the proposed method is presenting a novel supervised approach to unusual behavior detection (where many methods in this field are unsupervised) which demonstrates significantly improved results.
Chapter Preview

Hierarchal Decomposition For Unusual Fish Trajectory Detection

Coral reefs are one of the most important natural environments which should be monitored to understand the environmental effects caused by global warming, pollution and so forth. Investigating such environments needs long-term monitoring and automatic analysis, although the traditional way is manual processing which is very labor intensive and time consuming.

Analyzing fish behavior is useful to detect environmental changes as fish behavior reflects environmental conditions. This analysis can be made by extracting the change in behavior pattern of fish or by finding abnormal behaviors (Beyan and Fisher, 2013). For instance, by analyzing the behavior of fish hovering over coral, the health of coral can be determined.

There are many video surveillance systems to observe fish behavior. The most well known way to analyze fish behavior is using video recordings where the camera is capturing the fish trajectories from a fish tank or in an aquarium (Papadakis, Papadakis, Lamprianidou, Glaroulos, & Kentouri, 2012). Diving to observe using photography, hand-held video devices and optical systems are techniques that have been used to investigate fish behavior in natural environments. Acoustic systems, echo-systems and sonar have been used as well (Graham, Jones, & Reid, 2004). Alternatively, casting nets in the ocean and net casting with acoustic sensors are also popular to observe fish and determine their abundance (Spampinato et al., 2012). However, methods such as diving and net casting are not very suitable as they cause unusual fish behavior by frightening the fish. Moreover, with those approaches it is hard to capture huge amounts of data and to do long-term monitoring (Spampinato et al., 2012). In recent years, as digital video recording systems become cheaper, collecting data in natural underwater environments with a fixed camera set up which is continuously recording underwater videos has become possible (Boom et al., 2014). Such a system results in massive amounts of underwater video although automatically and accurately analyzing data is still a challenging problem. At this point, computer vision techniques and pattern recognition methods could play an important role in analyzing the fish behaviors using underwater videos.

In the computer vision area, behavior understanding studies can be classified into two categories:

  • Activity recognition,

  • Unusual behavior detection (Piciarelli, Micheloni, & Foresti, 2008).

When the number of possible behavior models in an uncontrolled and uncooperative real-world is considered, activity recognition is very challenging as the system needs a definition of each activity (Piciarelli et al., 2008). As fish are usually not goal-oriented and make erratic movements due to water currents, the complexity of the movements increases and makes encoding the behaviors into activities very challenging. On the other hand, unusual behavior detection and analysis has become popular in recent years. To detect unusual behaviors, the system generally does not need any prior knowledge about the behaviors. The unusual behaviors are generally defined as outliers or rare events and are detected in an unsupervised fashion (Anjum and Cavallaro, 2008; Jiang, Yuan, Tsaftaris, & Katsaggelous, 2010).

Complete Chapter List

Search this Book: