Recording and analysing animal movement patterns is invaluable not only for biological research, but also for evaluating and assessing ecological and economic issues [1–6]. To avoid misunderstandings, we would like to frame our definition of tracking as follows: Waypoints, i. e. coordinates indicating the position of the object to be observed can be acquired anywhere along its route or line of travel, apart from systemic exceptions (e.g. obstacles like trees for harmonic radar).
There are various methods to track animals. In general, a tracking method must meet different requirements to be used in natural or artificial environments. In contrast to an artificial environment, most parameters are not controllable in a natural environment (e. g. the background) and recording of more than a few waypoints of a route or line of travel is still challenging for small animals (<100 g) [6–8]. In the field, many different techniques utilizing different signals, e. g. electromagnetic- and acoustic waves, are used to track individuals. However, to the best of our knowledge, methods for tracking individual insects are limited to electromagnetic radiation. Thus, the focus of the manuscript lies exclusively on methods that use electromagnetic radiation.
As mentioned before it is still difficult to record movement patterns of small animals like honey bees — even if they are tagged [5, 6]. Honey bees are considered to be strong flyers capable of coping with payloads representing 20% of their body mass without any indication of an altered flight performance . For flying animals in general, only attached items that weigh less than 5% of the body are assumed not to affect flight performance — "5% rule“ — even if this assumption is not based on a very large data base [7, 10, 11]. Honey bees and butterflies (Aglais urticae, Melitaea cinxia, Agrotis segetum) mark the bottom end of the technical possible range of trackable flying animals. So far smaller flying insects can only be recorded in the field as assemblages [6, 13, 14].
The high demand for technical solutions to track small insects has led to technical progress. Thus, transmitters and transponders (active, semi-active, and passive), with which animals were equipped for tracking, became smaller and lighter over time. [7, 15–17]. In general, passive transponders (without their own power source) are much lighter than active and semi-active transponders and transmitters (which have their own power source).
The newest generation of battery powered active radio frequency (RF) identification tag weigh approximately 95 mg and reach an operational distance of about 1 km with a 10 cm whip-antenna . However, these transmitters have not been tested for their suitability to track animals yet. Although this approach is very promising, the length of the antenna is likely to be problematic [6, 12]. It might be reasonable to increase the frequency if technically possible, even if the lifespan — currently 16 days — of the tag might be reduced due to a higher power consumption and a higher susceptibility to interference. Another drawback of this new approach might be that it is based on complementary metal-oxide-semiconductor (CMOS) technology where the unit price extremely depends on the number of pieces produced. Finally, although 95 mg is a huge step forward as it represents a weight reduction of over 50% compared to other systems of this kind, it still adds about 80% of the weight of a honeybee. Such high payloads most likely affect energy consumption and flight performance [12, 14, 18–20].
A new unique and very promising approach utilizes a batteryless active radio frequency (RF) transmitter (which weighs 30 mg to 80 mg) with a small stationary receiver. Instead of having a battery it harvests the required energy (for the transmitter) via an piezoelectric harvester. The authors showed that their system works with an operational distance of about 10 m. Furthermore, they autonomously tracked a hand held battery equipped transmitter over a distance of 50 m with a drone. Thus, in near future it could be the first non-stationary tracking device mounted on a multicopter with which tracking of insects would be possible . According to their own statement the lightest tags (30 mg) are not easy to produce, which results in a high reject rate. It is possible that this method will soon be superior to radar technology in many respects, but at least in one respect it is not - the transmitters weigh at least twice as much as conventional radar tags [5, 14, 21].
To our knowledge all devices used in the field for passive tags are stationary ones with a small detection range of only a few centimetres up to some meters. The only exception is the harmonic radar system that can reach distances up to 1 km [6, 21–23]. Harmonic radar tags require an antenna which is between 12 mm to 16 mm long. Despite the fact that, due to the frequency, the antennas used are relatively short, various reviews of tracking methods point out that this is still a disadvantage of the method which might influence the behaviour of the animals under investigation [6, 12, 24].
Due to the utilized wavelength, harmonic radar tracking only works in a flat terrain without obstacles like trees or bushes [5, 25]. Another drawback is the low resolution in space (±2.5 m) and time (≥3 sec). Thus, it is not possible to distinguish between individuals if they meet or fly past each other within a small distance [6, 25]. However, for bumblebees Riley  found a mean flight speed of 7.1 ms-1 — in windless conditions. This means that a straight flight starting at the center of the largest possible area to be observed with a harmonic radar (radius<1000 m) is covered in less than 1.2 min. Thus, all prior stationary devices — even the harmonic radar — are just a peek inside the movement patterns of flying animals if they are not central place foragers.
There are two ways to address these shortcomings: a) by using many stationary devices or b) by tracking individuals with a non-stationary device. Since movement in space takes place in at least two dimensions — under the very strong simplification that the flight altitude is negligible, the number of stationary devices required for surveillance of a certain study site increases proportionally to its size. Therefore, the use of passive tags with a short detection range — not to mention the costs — is logistically hardly manageable, simply due to the above named relation. Thus, even if radio frequency identification (RFID) tags with a range of about 1.5 m and a weight of about 10 mg will be available in the near future , they will not be suitable for experiments comparable to those with harmonic radar. Consequently, in the foreseeable future, only b) remains as a feasible solution. One can easily imagine that if the mobile unit (i. e. the receiver or transceiver) would be airworthy — e. g. by mounting it to a drone, this would be a great solution. But it should be noted that the more thrust (i. e. the heavier the flying object) is required, the greater the downwash and therefore the distance to the observed object required for keeping it under a defined level of influence. On the transceiver/receiver’s side, there seems to be only one clear solution, to make it mobile.
A yet unanswered question for tracking is whether the size or the radio frequency band of the transponder/transmitter is more important. In general, when using radio frequencies for communication, it should be noted that the frequency used has a decisive influence on the susceptibility of the system to interference (i.e. disruptive effects of the environment)[6, 12, 27–29]. The susceptibility to interference from water and reflections correlates positively with the frequency used [27–29]. This is particularly important when tracking animals in the wild, as the signal has to travel through the (natural) environment.
However, since the optimal antenna length is proportional to the wavelength, lower frequencies require longer antennas which affects tag (i.e. transmitter or transponder) size  — consequently, miniaturisation has its price. Interestingly, the harmonic radar uses the lightest tags (with super high frequency (SHF) band) of any tracking method known to us. From the fact that harmonic radar sets the standard for tracking of flying insects for more than two decades, [9, 21, 22, 25, 26, 30], so that new methods are always compared to it [15–17], we conclude that size is more important.
In summary, to be able to track insects over greater distances than 1000 m the tracking systems must be mobile and the tags used must be small and light to fulfil the 5% rule. There is currently no RF tracking solution that serves both since either the used RF tags are simply not small and light enough or making the receiver/tranceiver unit mobile for autonomous tracking is not possible. Therefore, it can be stated that RF tracking methods can currently not be considered as an uncompromisingly applicable solution.
In the following section we will look at visual tracking methods. Throughout the manuscript, we restrict ourselves to real-time (RT) capable systems so that they can be used as non-stationary devices in the first place. We think that stationary devices are too big a logistical challenge for the same reasons given above for RFID systems. For image-based object tracking, the frame rate and the evaluation speed (i. e. delay) is the limiting factor since a loss of the object from the field of view (FOV) usually can not be compensated. Since several years even consumer drones equipped with a vision system are capable of actively tracking and following physically untagged objects autonomously.
However, tracking small untagged objects like flying insects is not possible with a drone yet. Tracking of insects has been done for a long time, for small routes or lines of travel, with stationary devices under laboratory conditions. Nowadays, even real-time tracking of multiple small objects like flies is possible with a delay that is small enough to manipulate sensory feedback . However, all such approaches have an average delay in excess of 40 ms [31–33]. This sounds small, but if we come back to the average flying speed of bumblebees (v), this translates to an inaccuracy of more than 28 cm. To achieve a smaller tracking error, one could slow down the flight speed for example by reducing the size of the flight arena to a relatively small one.
However, when tracking a free-flying insect with a mobile-camera, the position and orientation of the camera would have to be updated very frequently so that the small object to be tracked does not leave the FOV of the camera, as this would mean the end of tracking. In general it can be said that a high update rate of the actual insect’s position with low latency in determining its position increases the probability that an animal can be faithfully tracked. A high sampling rate can be seen as a buffer for tracking errors, since it means that more waypoints can be recorded over the same route or line of travel.
Although there are real-time (RT) visual systems to record trajectories of flying insects, they are all only capable of doing so under known, slowly changing environmental conditions, e. g. constant background. Furthermore, none of them is hard real-time capable. But the control system design of a non-stationary sensor system is made substantially easier if variability of detection lag can be ignored. This condition is met, for example, if the system is built from hard real-time components. Otherwise, the control system must have a way of dealing with sensor data of unpredictable timing. Fortunately, our system is a hard real-time (HRT) system without being restricted to slowly changing backgrounds, which makes it an ideal candidate for non-stationary tracking.