Question 1

What are the steps typically involved in a given buoy-based monitoring project?

Accepted Answer

Question 2

What is the detection range for the buoy-based vision system?

Accepted Answer

Spoor’s AI software detects, tracks and recognizes birds based on visual camera recordings. The ability of the AI algorithm to detect birds and calculate their positions, are mainly affected by bird size, weather conditions and camera hardware.

The below illustration gives examples for a standard, marine-grade 4K camera. Please note that the range varies depending on camera hardware.

Question 3

Is Spoor approved by regulators?

Accepted Answer

Technologies and methods for bird detection are usually not directly approved by regulators in a formal process, but rather indirectly when regulators approve the use of technology in specific monitoring programmes.

Spoor has been used for monitoring in three regulatory accepted monitoring programmes in Norway. In a research project with the British Trust for Ornithology, an expert group has consisted of representatives from regulators in the UK.

Spoor has in-house expertise on regulatory affairs. In a scoping process, Spoor works together with the customer to inform the local regulator, and shape the monitoring programme so that it comply with the formal requirements.

Question 4

Can Spoor be used to track migration?

Accepted Answer

Spoor can be deployed to capture valuable data relating to migration:

Whether migratory species are flying through the area at altitudes that can may coincide with turbines.

The weather conditions that coincides with observations of migratory species

If a migratory bird uses the site for resting, nesting, foraging or other activity, this data will be captured and analysed

The interaction between a migratory bird, and turbines

Defining migration as the regular seasonal movement of birds, often north/south along a flyway, with the purpose of moving from areas of low or decreasing resources to areas of high or increasing resources (mainly food).

Spoor's method of camera-based monitoring is an example of Eulerian sampling; monitoring birds within a fixed volume. The Spoor application has been tailored to monitor spaces for wind farming. It can help answer the question of whether such a space is coinciding with a migratory route, within the altitudes that will be occupied by wind turbines. It can also help answer whether the site is used by migratory bird species for other purposes; like breeding or foraging.

The heights most interesting for wind farm sites are up to ~300m offshore and ~200m onshore, as these are the altitudes that turbines will occupy. In comparison, long-distance migrating birds may fly at altitudes from 1500 m up to 6000 m. Lagrangian methods that focus on turbine-height altitudes will naturally not register birds flying at higher altitudes. Keep in mind that the monitoring range also depends on the size of the species in question. Smaller birds have a shorter range than larger birds.

Question 5

What equipment and hardware is required to run the Spoor system?

Accepted Answer

Spoor's vision systems are based on commercially available surveillance cameras as sensors.

Spoor assists customers by designing a setup that caters to the conditions and purpose of the site in question, including a recommended camera type.

In general, these are the elements needed in order to run the Spoor bird detection, tracking and identification software:

High-resolution outdoor surveillance camera

Including a secure camera mount that minimise mechanical vibrations

Certain camera types requires an edge computer in order to record and stream video

Cable(s) for connecting the camera to power and internet connection

Usually a PoE cable and PoE adaptor

If power and/or sufficient internet connection is not readily available at the camera vantage point, additional equipment is needed. In the design phase of a project, Spoor will suggest an appropriate setup that provides this.

Question 6

Where has Spoor been deployed?

Accepted Answer

Per Q3 2024, Spoor has been deployed on 16 sites on- and offshore, spanning Scotland, France, Germany and Norway.

Question 7

Can Spoor monitor bats?

Accepted Answer

Spoor has no commercialized product yet designed to solely monitor bats, but are providing thermal-based detection and tracking vision systems that can capture bats.

Question 8

What species can Spoor detect?

Accepted Answer

There is no limit on which species can be detected. As long as a bird appears on the camera stream for more than 1 second and measures over 8 pixels in the image, Spoor's algorithm will detect and track the bird.

In addition to detecting and tracking, Spoor works with algorithms that automatically recognises the species. These algorithms requires higher-resolution videos than what is required for detection and tracking. When such high resolution video is not available or not applicable, Spoor's in-house ornithologist manually classifies species. Depending on the purpose, this can be done for a certain time period, or for a randomised or selected subset of all the bird detections.

Per Q3 2024, Spoor has tracked 77 different species, both terrestrial and marine.  The list includes Golden Eagle, Rough-legged Buzzard, Red Kite, Whooper Swan, Western Caperacillie, Northern Gannet, Little Gull, Black Woodpecker, Peregrine Falcon, Great Grey Shrike, Bohemian Waxwing, Eurasian Curlew, Black-legged Kittiwake, Black-headed Gull, European Herring Gull, Northern Fulmar, Arctic Jaeger, Common Tern, White-tailed Eagle, Hen Harrier, Western Marsh Harrier, Northern Goshawk, European Starling and Northern Lapwing.

Question 9

How does Spoor compare to traditional methods?

Accepted Answer

The "traditional method" is usually defined as human observers, with laser-rangefinders to determine a bird's flight height. There are certain limitations of a human observer that an automated system like Spoor can overcome:

DURATION: A human is on-site for a very limited time.

VERIFIABLE: The observations are not verifiable, so it is the notes of a single person that is used.

CONTINUOUS: An ornithologist takes breaks, and gets cold and goes home.

COMPLETE: An ornithologist is subject to human error. The human can be inattentive or occupied with taking notes.

REPRESENTATIVE WEATHER CONDITIONS: The ornithologist only observes in nice weather.

COGNITIVE LOAD: The human has a cognitive limit.

Comparing Spoor to traditional methods of bird monitoring

Spoor conducts regular quality verifications by comparing Spoor AI results to the field observations of a human observer, usually a trained ornithologist. This is done at different sites to ensure quality is measured across various environments.
In this method, the human observer visits a site and manually records bird detections within a field of view that match the field of view of the camera used for Spoor data collection.

The bird detections are noted down with information on: time of the bird entering the field of view and exiting the field of view, the bird's trajectory through the field of view, and its species. If the verification focus is spatial mapping, the human observer records flight height and distance from the observation point. In order for this to be precise, the ornithologist uses an instrument like a laser rangefinder or binoculars with rangefinders. This data set represents a "Ground Truth"; information that is known to be real or true, provided by direct observation and measurement (i.e. empirical evidence).

The Ground Truth data set is subsequently compared with the Spoor AI detections. Because the human observer is also subject to errors, some birds are missed by the human observer but are detected by the Spoor AI. The number of birds missed by humans are determined in the comparison, and added to the Ground Truth data set. This yields a "more true" ground truth than would be achieved by only taking the manual observations as the ground truth. In this way, it serves as a cross-verification of both the Spoor AI and the manual observer, allowing for assessing quality and levels of error of both methods.

Spoor's system is as good as an ornithologist

Based on 5 field trips on 4 different sites during 2023, Spoor has compared bird counts by a trained ornithologist, and the system's bird counts. The results showed that Spoor is as good as the ornithologist. Due the inherent self-learning properties of the AI combined with Spoor's focus on quality assurance, these results will continuously improve and surpass the ornithologist in near future.

Recall measures the proportion of birds found by the system

Within machine learning and AI, recall is the measure of how many of the present birds the AI manages to detect. Recall is calculated by dividing the number of detected birds by the number of present birds.

The average recall for Spoor and the ornithologist was 70% and 71%, respectively. Weighted by duration, the Spoor weighted average recall was 75% with a standard deviation of 12 percent points and the ornithologist weighted average recall was 76% with a standard deviation of 9 percent points.

Analysis of the results showed that the AI system missed occasional birds that were part of flocks. The birds that the ornithologist missed were to a larger degree single birds that appeared alone. The reason why a manual observer misses birds is mainly due inattention, weariness or focus on other tasks like recording a previous observation.

Question 10

Has Spoor's method been verified?

Accepted Answer

In addition to Spoor's internal quality assurance, two formal verifications have been conducted:

An offshore system verification that were conducted as part of a report to regulators. Bird counts by ornithologists in a vessel was compared to bird counts by the Spoor-system. The results were a 100% overlap between the ornithologists and the Spoor-system, within the camera field of view, but the sample was too small due to temporal limitations of the ornithologists.

A research project in collaboration with Vattenfall and the British Trust for Ornithology, which is to be published in 2024. The project also involved an expert group consisting of members from regulators in the UK, and NGO representatives.

One part of the research was an evaluation of the ability to correctly determine the bird position in space. This was done by tracking a drone, and comparing the flight trajectory tracked by the Spoor-system, and the drone's GPS track. This was conducted on land.

Another part of the research was comparison of the traditional method of human observations with laser-rangefinder, and the Spoor-system. This was conducted on birds in an offshore wind farm.

A clear conclusion from the research project is that all methods have error sources. Since the error sources are different for each method, there is no one method that can be deemed better than others and that can serve as "ground truth". In practice, human observations are often used as the standard without recognising the error sources and limitation of this method. With a human observer, the measurement volumes are unknown, what direction the human is watching, which birds they miss and why, what the errors are for the laser-rangefinder, what periods they are inattentive, etc. It is therefore of outmost importance to understand the limitations and benefits of each method individually, rather than taking the imperfect method of human observations as the ground truth.

Question 11

What detection ranges can be achieved?

Accepted Answer

The distance of which a bird is visible is important because it dictates the space and area that can be covered with one camera.

Larger objects are detectable over longer distances than smaller objects. This applies to devices like visual spectrum and thermal cameras, radars, telescopes, and the human eye. Intuitively, we know that the leaf of a tree is a small object that is only visible over a few metres with the human eye, while the moon is a very large object that is visible over hundreds of thousands of kilometres. For this reason, large birds are detectable over a longer distance than smaller birds. When
discussing detection range, bird size needs to be taken into account.

The maximum detection range also depends on the properties of the sensor equipment. For a mid-grade camera with a standard lens, a bird with a wing span of 150 cm (like a Great Black-Backed Gull) can in theory be detected up to 2 km away, while a bird with a 25 cm (like a Meadow Pipit) can be detected up to 350 m away. Increasing the lens zoom capabilities and/or the camera resolution will yield greater ranges. By improving the camera equipment, e.g. by upgrading the lens, the detection range can be increased to several kilometers.

Spoor's engineers and environmental experts will suggest a recommended setup that caters to the specific site conditions, budget and purpose of each project.

Question 12

What are the requirements in terms of cables?

Accepted Answer

All the cameras need a Minimum grade CAT6 outdoor shielded ethernet cable with power for the outside connection. This cable may have to be marine cable if the cameras are exposed to waves. In addition, our system needs CAT6 indoor internet cable to transmit the data and CAT6 indoor power cable.

Question 13

Does Spoor personnel require physical access to the site?

Accepted Answer

Spoor has full remote control of the Spoor software, and of most camera interfaces that controls zoom levels.

In situations where Spoor, or a Spoor subcontractor, is responsible for the installation or maintenance of a camera system, site access will be necessary. In these cases, Spoor will work with the site owner to fulfill the appropriate health and safety requirements and training.

Question 14

What is a sufficient internet connection?

Accepted Answer

Internet connection is typically facilitated by either 4G, fiber or satellite. The connection is used both for transmitting data from site to the Spoor dashboard, and for remote monitoring of camera and potential data storage systems.

Full data transmission: 25 Mbps/camera minimum

Part data transmission: Below 25 Mbps/camera. This requires Edge Computing on-site storage and manual data retrieval.

Low/No data transmission: Some sites do not provide connectivity, and Spoor will propose a hardware design that can cater for this.

Question 15

How are cameras mounted?

Accepted Answer

Different options are possible, and depend on the site properties, the chosen cameras and the project’s requirements (weather conditions, species of interest…). Ideally, cameras are mounted in a way that minimizes mechanical vibrations, provides sufficient leeway to orient each camera, ability to access for maintenance, and ability to attach securely.

Spoor is currently working with three alternatives for camera mounting:

Circular support from axis which permits attaching Axis cameras to poles. There are two different models regarding the diameter of the pole: from 100 to 410 mm or from 50 to 150 mm.

Railing support to attach to a safety fence: Spoor is not able to provide the dimension of these rails. It has to be adapted to the safety rail. A subcontractor will design and fabricate camera mounts

Magnets: Already used before in offshore turbines. The magnets are directly attached in the support of the camera

Question 16

If a site cannot provide internet connection, how frequently do we need to collect data manually?

Accepted Answer

If the data needs to be stored on-site because it cannot be transferred via network/internet, the data needs to be collected physically. The frequency of physically retrieving the data depends on the video resolution (higher resolution/better quality have larger file size) and of the data storage capacity of the selected device.

Example 1:
1 Axis LE camera recording 12 h/day with data storage of 1 TB (Maximum size for SD card):

With this setup, the data storage will reach 80% capacity within 14 days.

Example 2:
4 Axis Dome cameras recording 6 hours/day and a NAS using 4 SSD disks of 8 TB:
With this setup, the data storage will reach 80% capacity within 6 months.

It is worth noting that it is a big logistics exercise to collect data manually. Spoor strongly recommends using a network for data transfer whenever possible.

Question 17

What are Spoor's recommendations regarding camera placement?

Accepted Answer

South to north camera angles where possible in northern hemisphere (to reduce sun strike issues on dirty or salt spray or rain drop affected lenses)

The shortest turbine to turbine distance possible, (minimizes vibration problems which degrades images)

The allowable camera positions on the turbine maintenance platforms (mounting positions that will not conflict with turbine working areas like access gates and cranes)

Dominant wind direction - this will blow dust, salt spray, rain onto the camera lenses if cameras are facing into this

Aim for the most stable, vibration free camera mounting options (eg. heavy structures unlikely to conduct vibrations)

Access to reliable power and to VPN/fibre network (within recommended range of cables without data loss) is also recommended, but Spoor is able to provide a standalone hardware rig for power supply and network connection for sites that lacks infrastructure.

Question 18

How is the camera system connected to power?

Accepted Answer

Each camera is connected with a PoE (Power over Ethernet) cable which is a cable that transmits both electricity and data. Using a PoE reduces the number of cables needed.

Example 1: Single camera with internet connection
For a single camera, the PoE cable will be connected to a PoE injector which is a device that splits the power and the data. On the other side of this device you will have a power connection through a “normal” plug and an internet connection through fiber.

Example 2: Multiple cameras with a storage system
The PoE cable of each camera is connected to the switch of the system. This switch is connected to a power access ( the customer UPS or directly a normal plug) and to the storage system

Note: For post-construction monitoring where cameras are mounted on turbines, the power connection is ideally from a connection that won’t be unplugged by visiting technicians when doing maintenance work.

Question 19

What are the power requirements of a typical camera system?

Accepted Answer

The power required depends on the camera and camera housing needed, and whether limitations in internet bandwidth require additional data storage and on-site storage with manual data retrieval.

Below are examples for two different setups, but please note that this is a case-by-case calculation and that the camera selection and potential data storage and processing equipment will decide the power consumption at your site.

Example 1:
Rig: 2 Axis LE on site with full internet connection
Power consumption: ~30W

Example 2:
Rig: 5x Axis LE on site with limited internet connection; requiring on-site storage and manual data retrieval.
Power consumption: ~ 140W

Frequently asked questions

Buoy-based vision system

General

Hardware and installation