<< 26/11 Mumbai attacks in India demonstrated that the threat from sea is real. There is thus a need to maintain continuous vigil at the waters and detect threats before they cross the borders. One of the country’s premier DRDO laboratory, Centre for Artificial Intelligence and Robotics (CAIR) is working in this direction… >>
The ever increasing threat from the sea has necessitated strengthening of our coastal security mechanism. There is an urgent need to keep our vast maritime zones under effective round-theyear surveillance by utilising all the assets at our disposal.
To appreciate the quantum of problems faced by the security agencies in maritime domain, one needs to visualise the mercantile traffic transiting through the Indian Ocean Region (IOR) or approaching the Indian Coast. Approximately one lakh ships transit through the IOR annually. This includes 66 per cent of the world’s oil traffic, 33 per cent of the bulk cargo and 50 per cent of the container shipments. Within this world wide flow of maritime trade, Indian maritime trade forms a significant part of the IOR flow. With 618 ships and 6.5 million (approx.) Gross Register Tonnage (GRT), the Indian foreign trade transacted over sea is 90 per cent by volume and 77 per cent by value.
The high traffic density in the IOR and the traffic that approaches to the Indian coast is further increased due to the presence of fishing vessels and dhows. India has more than three lakh registered fishing vessels with an approximate annual catch of eight million tons. Also there is a large amount of dhow traffic. The high density of merchant traffic in our area of interest has made vessel identification one of the biggest problems in maritime reconnaissance. This is further complicated because of factors like maritime terrorism, piracy, trafficking (drugs, human) and gun running.
To safeguard the maritime domain and specifically the approaches to the Indian Coast, it is important that the various security agencies, in the effective discharge of their responsibility in respect of maritime and coastal security, must have enhanced Maritime Domain Awareness (MDA) of the IOR. To meet the same, Indian maritime stakeholders have entered/ are entering into partnerships for information exchange using online information exchange portals with regional navies having substantial presence in the IOR or the countries straddling the entrance gateways of the Indian Ocean. Other multiple initiatives are underway to enhance the situational awareness and strengthen the coastal security like National Automatic Identification System (NAIS) Chain, Long Range Identification and Tracking (LRIT) and Static Sensor Chain by Bharat Electronics Limited (BEL).
All these initiatives cater for commercial vessels or the vessels with capacity of 300 GRT and above. The identification and tracking of traditional fishing boats and sub 20 meter vessels are not accounted in these initiatives. To fill this gap, Indigenous AIS (IAIS) is being developed which targets vessels with size < 20 m or gross weight < 300 GRT. The IAIS is a Mobile SATCOM Services (MSS) based tracking system fitted on vessels for identifying and locating vessels by electronically exchanging data with SATCOM based AIS Management System (SAMS) running at a central hub. SAMS in turn passes this information to a Central Control Station (CCS). Information such as vessels unique identification, position, course and speed are displayed on a screen at the CCS. IAIS information helps to isolate unknown or suspect vessel by correlating it with radar information. SAMS also provides value added information from Indian National Centre for Ocean Information Services (INCOIS), regarding Potential Fishing Zone (PFZ), sea states and ocean weather information, etc. to the registered vessels installed with the IAIS terminal.
In view of the above requirement, maritime situation awareness system is being developed for generating a combined synthetic operational picture with feeds from various sensors like radars, electro optic sensors, MET sensors, AIS, IAIS and underwater sensors. This facilitates sharing of information for the generation of Common Operational Picture (COP) among the various nodes across the hierarchical echelons and among various maritime stakeholders. The system enables the maritime security agencies for early detection and monitoring of unregistered or suspicious vessels operating in a given controlled area of the country’s coastal waters with minimal operator’s intervention.
GIS for MDA
GIS is a crucial component of any maritime system since it enables the user to view the COP, assess the current operational scenario, grasp the situation and quickly act upon it. Here the main function of the GIS is to display tracks and alerts with S57/ S63 charts as underlay using S-52 display libraries. The other functionality of GIS includes filtering, object selection, overlay management, spatial queries and geodesic and geometrical calculations.
The design of the GIS system decouples the rendering module and other GUI components from the GIS engine. This gives the flexibility of integrating the GIS engine to any other suitable rendering system and GUI packages. The GIS engine performs the core function of managing layers and features, datum, projections, spatial query, etc. The rendering system displays the GIS information to the user on a suitable graphical user interface. The important modules are shown in the Figure 1.
Figure 1: GIS system modules
The application is designed to handle a large number of targets captured by sensors and communication systems. Even though the system is capable of processing large number of targets in the background, displaying targets every second in the graphics view is a compute resource intensive task. Hence multiple performance enhancement techniques have been employed to meet the performance requirements while processing, managing and displaying a large number of tracks on the GIS. Further, area based and speed based clusters have been implemented to improve the visualisation and make the presentation more user friendly. The following challenges were observed:
The refresh rate of tracks perceived by the user is different and higher than the track update frequency of various sensors. To achieve the continuous visualisation, tracks are extrapolated based on the previous parameters such as speed, course, last-position and time. It was observed that at higher zoom out, the tracks with very low speeds do not change the position significantly enough and do not require a redraw.
Speed based clustering
Speed based clustering of the tracks is envisaged to reduce the overhead of computing too many extrapolations. To reduce the extrapolation overhead of CPU, we maintain a cluster of tracks based on speed of the tracks. For example, a track with higher speed such as air tracks needs to be extrapolated at faster rate whereas a submarine track which runs at slower speed are extrapolated at slower interval without any perceptible loss of accuracy. Further, even if we use a highly accurate geodesic algorithm for the extrapolation of track, the accuracy is lost during worldcoordinate to device-coordinate conversion. Hence, for the extrapolation of tracks, two types of geodesic algorithms are to be used:
Spherical extrapolate: This extrapolate uses the spherical earth model which consumes less CPU resources, but it provides reduced accuracy. When the required level of detail is less (that is, if the tracks are to be extrapolated only for drawing on the GIS), then the spherical extrapolate is used.
Figure 2: Track display after zoom In
Figure 2b: Track display after zoom out
Ellipsoid extrapolate: The ellipsoid extrapolate is used for highly accurate calculations based on the WGS84 (World Geodesic System) ellipsoid model, but it consumes more CPU resources. Hence, whenever a user is examining the tracks in detail (which are fewer in number), the ellipsoid extrapolate algorithm is used.
However, periodically the ellipsoid extrapolate algorithm is applied to all tracks to remove the accumulated error generated by the spherical extrapolate.
Track symbol performance
The system draws each target on the screen at frequent intervals, that is, refresh rate defined by the user. For example, every second for surface tracks and 100 milliseconds for air tracks. The track symbols are vector images consisting of, on average, 10-15 lines. The symbols are to be represented on the screen with a correct orientation with respect to its course; so all the symbols need to be rotated before displaying them. This introduces an additional overhead of rotating the track symbols. Similarly, displaying text labels for each track introduces a lot of overheads. This reduces the overall system performance.
Look-up table: The rotation of track symbols requires multiplication by trigonometric functions. Instead of rotating the symbols at run time, the system stores the rotated symbols between zero and 359 degrees in a lookup table in the memory. Similarly, the trigonometric functions that are extensively used by extrapolation function are evaluated for various parameter values and these computed values are stored in an indexed memory.
If within a small geographical area, the concentration of tracks is high, then the display becomes cluttered and it is difficult to distinguish the tracks. The display becomes further cluttered due to rendering of the track history and track labels. If the number of tracks is very large then this overlapping is so extensive that the user is not able to identify tracks. Therefore, it was noted that even after carrying out resource intensive processing to display all tracks, the user does not get any benefit from this visualisation.
Keeping in view the above observations, the following design is being used to handle millions of tracks efficiently:
Area based clustering: To reduce the cluttering on the screen and display the large number of tracks in a small geographical area, we use the technique called clustering. In this technique, the tracks are grouped into a single icon when the number of tracks within an area is greater than a certain threshold. This benefits the user by removing the excessive clutter and displaying the information in a more comprehensive manner. The clustering also reduces the number of objects to be drawn on the screen resulting in a drastic improvement in performance. With experimentation and prototyping, we have found that by using clustering we can extrapolate and draw one million tracks within 1.5 seconds. This includes the cluster creation time also. This result was calculated by using only area based cluster. The combination of both area and speed based cluster further improves the performance.
Figure 3 : Track display after track clustering
Future work – Decision support system
At a particular node, the track data received from various sensors (radar, sonar and comint, etc.) is processed and verified against data inputs from the AIS/ IAIS receivers for identification. The data from all these sensors is fused to generate the situational awareness picture. The situational awareness picture generated at a node is shared among the multiple nodes to generate the COP. Given the vast amount of data which is generated using multiple sensors, it is humanly impossible to detect the suspicious objects on COP. Hence the system uses various self learning artificial intelligence (AI) algorithms to detect the variance from the regular patterns and detect the anomalies/ irregularities that occur in the domain. The precision for the alarms raised is improved using the information extracted by investigating the behaviour over both space and time for individual objects.
The system maintains a knowledge base of the normal course of events extracted from the track data received over a period. System operates autonomously, finds objects that do not behave according to what is considered normal, raises the alarm for the detected anomalies and highlights the same to the user to enable effective and timely action using the limited human resources optimally.