Insect-Inspired Estimation of Egomotion

Moving object detection based on bioinspired background subtraction

Flying insects rely mainly upon visual motion to detect and track objects. There has been a lot of research on fly inspired algorithms for object detection, but few have been developed based on visual motion alone. One of the daunting difficulties is that the neural and circuit mechanisms underlying the foreground-background segmentation are still unclear. Our previous modeling study proposed that the lobula held parallel pathways with distinct directional selectivity, each of which could retinotopically discriminate figures moving in its own preferred direction based on relative motion cues. The previous model, however, did not address how the multiple parallel pathways gave the only detection output at their common downstream. Since the preferred directions of the pathways along either horizontal or vertical axis were opposite to each other, the background moving in the opposite direction to an object also activated the corresponding lobula pathway. Indiscriminate or ungated projection from all the pathways to their downstream would mix objects with the moving background, making the previous model fail with non-stationary background. Here, we extend the previous model by proposing that the background motion-dependent gating of individual lobula projections is the key to object detection. Large-field lobula plate tangential cells are hypothesized to perform the gating to realize bioinspired background subtraction. The model is shown to be capable of implementing a robust detection of moving objects in video sequences with either a moving camera that induces translational optic flow or a static camera. The model sheds light on the potential of the concise fly algorithm in real-world applications.

Sparse Optical Flow Implementation Using a Neural Network for Low-Resolution Thermal Aerial Imaging

This study is inspired by the widely used algorithm for real-time optical flow, the sparse Lucas-Kanade, by applying a feature extractor to decrease the computational requirement of optical flow based neural networks from real-world thermal aerial imagery. Although deep-learning-based algorithms have achieved state-of-the-art accuracy and have outperformed most traditional techniques, most of them cannot be implemented on a small multi-rotor UAV due to size and weight constraints on the platform. This challenge comes from the high computational cost of these techniques, with implementations requiring an integrated graphics processing unit with a powerful on-board computer to run in real time, resulting in a larger payload and consequently shorter flight time. For navigation applications that only require a 2D optical flow vector, a dense flow field computed from a deep learning neural network contains redundant information. A feature extractor based on the Shi-Tomasi technique was used to extract only appropriate features from thermal images to compute optical flow. The state-of-the-art RAFT-s model was trained with a full image and with our proposed alternative input, showing a substantial increase in speed while maintain its accuracy in the presence of high thermal contrast where features could be detected.

An insect-inspired model for acquiring views for homing

Wasps and bees perform learning flights when leaving their nest or food locations for the first time during which they acquire visual information that enables them to return successfully. Here we present and test a set of simple control rules underlying the execution of learning flights that closely mimic those performed by ground-nesting wasps. In the simplest model, we assume that the angle between flight direction and the nest direction as seen from the position of the insect is constant and only flips sign when pivoting direction around the nest is changed, resulting in a concatenation of piecewise defined logarithmic spirals. We then added characteristic properties of real learning flights, such as head saccades and the condition that the nest entrance within the visual field is kept nearly constant to describe the development of a learning flight in a head-centered frame of reference, assuming that the retinal position of the nest is known. We finally implemented a closed-loop simulation of learning flights based on a small set of visual control rules. The visual input for this model are rendered views generated from 3D reconstructions of natural wasp nesting sites, and the retinal nest position is controlled by means of simple template-based tracking. We show that naturalistic paths can be generated without knowledge of the absolute distance to the nest or of the flight speed. We demonstrate in addition that nest-tagged views recorded during such simulated learning flights are sufficient for a homing agent to pinpoint the goal, by identifying nest direction when encountering familiar views. We discuss how the information acquired during learning flights close to the nest can be integrated with long-range homing models.

Insect-Inspired Self-Motion Estimation with Dense Flow Fields--An Adaptive Matched Filter Approach

The control of self-motion is a basic, but complex task for both technical and biological systems. Various algorithms have been proposed that allow the estimation of self-motion from the optic flow on the eyes. We show that two apparently very different approaches to solve this task, one technically and one biologically inspired, can be transformed into each other under certain conditions. One estimator of self-motion is based on a matched filter approach; it has been developed to describe the function of motion sensitive cells in the fly brain. The other estimator, the Koenderink and van Doorn (KvD) algorithm, was derived analytically with a technical background. If the distances to the objects in the environment can be assumed to be known, the two estimators are linear and equivalent, but are expressed in different mathematical forms. However, for most situations it is unrealistic to assume that the distances are known. Therefore, the depth structure of the environment needs to be determined in parallel to the self-motion parameters and leads to a non-linear problem. It is shown that the standard least mean square approach that is used by the KvD algorithm leads to a biased estimator. We derive a modification of this algorithm in order to remove the bias and demonstrate its improved performance by means of numerical simulations. For self-motion estimation it is beneficial to have a spherical visual field, similar to many flying insects. We show that in this case the representation of the depth structure of the environment derived from the optic flow can be simplified. Based on this result, we develop an adaptive matched filter approach for systems with a nearly spherical visual field. Then only eight parameters about the environment have to be memorized and updated during self-motion.

Motion as a source of environmental information: a fresh view on biological motion computation by insect brains

Despite their miniature brains insects, such as flies, bees and wasps, are able to navigate by highly erobatic flight maneuvers in cluttered environments. They rely on spatial information that is contained in the retinal motion patterns induced on the eyes while moving around ("optic flow") to accomplish their extraordinary performance. Thereby, they employ an active flight and gaze strategy that separates rapid saccade-like turns from translatory flight phases where the gaze direction is kept largely constant. This behavioral strategy facilitates the processing of environmental information, because information about the distance of the animal to objects in the environment is only contained in the optic flow generated by translatory motion. However, motion detectors as are widespread in biological systems do not represent veridically the velocity of the optic flow vectors, but also reflect textural information about the environment. This characteristic has often been regarded as a limitation of a biological motion detection mechanism. In contrast, we conclude from analyses challenging insect movement detectors with image flow as generated during translatory locomotion through cluttered natural environments that this mechanism represents the contours of nearby objects. Contrast borders are a main carrier of functionally relevant object information in artificial and natural sceneries. The motion detection system thus segregates in a computationally parsimonious way the environment into behaviorally relevant nearby objects and-in many behavioral contexts-less relevant distant structures. Hence, by making use of an active flight and gaze strategy, insects are capable of performing extraordinarily well even with a computationally simple motion detection mechanism.

Depth information in natural environments derived from optic flow by insect motion detection system: a model analysis

Knowing the depth structure of the environment is crucial for moving animals in many behavioral contexts, such as collision avoidance, targeting objects, or spatial navigation. An important source of depth information is motion parallax. This powerful cue is generated on the eyes during translatory self-motion with the retinal images of nearby objects moving faster than those of distant ones. To investigate how the visual motion pathway represents motion-based depth information we analyzed its responses to image sequences recorded in natural cluttered environments with a wide range of depth structures. The analysis was done on the basis of an experimentally validated model of the visual motion pathway of insects, with its core elements being correlation-type elementary motion detectors (EMDs). It is the key result of our analysis that the absolute EMD responses, i.e., the motion energy profile, represent the contrast-weighted nearness of environmental structures during translatory self-motion at a roughly constant velocity. In other words, the output of the EMD array highlights contours of nearby objects. This conclusion is largely independent of the scale over which EMDs are spatially pooled and was corroborated by scrutinizing the motion energy profile after eliminating the depth structure from the natural image sequences. Hence, the well-established dependence of correlation-type EMDs on both velocity and textural properties of motion stimuli appears to be advantageous for representing behaviorally relevant information about the environment in a computationally parsimonious way.

A μ analysis-based, controller-synthesis framework for robust bioinspired visual navigation in less-structured environments

Safe, autonomous navigation by aerial microsystems in less-structured environments is a difficult challenge to overcome with current technology. This paper presents a novel visual-navigation approach that combines bioinspired wide-field processing of optic flow information with control-theoretic tools for synthesis of closed loop systems, resulting in robustness and performance guarantees. Structured singular value analysis is used to synthesize a dynamic controller that provides good tracking performance in uncertain environments without resorting to explicit pose estimation or extraction of a detailed environmental depth map. Experimental results with a quadrotor demonstrate the vehicle's robust obstacle-avoidance behaviour in a straight line corridor, an S-shaped corridor and a corridor with obstacles distributed in the vehicle's path. The computational efficiency and simplicity of the current approach offers a promising alternative to satisfying the payload, power and bandwidth constraints imposed by aerial microsystems.

An insect-inspired bionic sensor for tactile localization and material classification with state-dependent modulation

Insects carry a pair of antennae on their head: multimodal sensory organs that serve a wide range of sensory-guided behaviors. During locomotion, antennae are involved in near-range orientation, for example in detecting, localizing, probing, and negotiating obstacles. Here we present a bionic, active tactile sensing system inspired by insect antennae. It comprises an actuated elastic rod equipped with a terminal acceleration sensor. The measurement principle is based on the analysis of damped harmonic oscillations registered upon contact with an object. The dominant frequency of the oscillation is extracted to determine the distance of the contact point along the probe and basal angular encoders allow tactile localization in a polar coordinate system. Finally, the damping behavior of the registered signal is exploited to determine the most likely material. The tactile sensor is tested in four approaches with increasing neural plausibility: first, we show that peak extraction from the Fourier spectrum is sufficient for tactile localization with position errors below 1%. Also, the damping property of the extracted frequency is used for material classification. Second, we show that the Fourier spectrum can be analysed by an Artificial Neural Network (ANN) which can be trained to decode contact distance and to classify contact materials. Thirdly, we show how efficiency can be improved by band-pass filtering the Fourier spectrum by application of non-negative matrix factorization. This reduces the input dimension by 95% while reducing classification performance by 8% only. Finally, we replace the FFT by an array of spiking neurons with gradually differing resonance properties, such that their spike rate is a function of the input frequency. We show that this network can be applied to detect tactile contact events of a wheeled robot, and how detrimental effects of robot velocity on antennal dynamics can be suppressed by state-dependent modulation of the input signals.

Biomimetic compound eye with a high numerical aperture and anti-reflective nanostructures on curved surfaces

Biomimetic compound eyes with a high numerical aperture on a curved surface were successfully fabricated by intelligent integration of traditional top-down and bottom-up micro- and nanofabrication methods together. In addition, the new hybrid micro- and nanofabrication method allows us to fabricate the antireflective nanostructures on each ommatidium to increase its vision sensitivity by improving the light transmission. The fabricated compound eye was optically characterized and was shown to have a numerical aperture of 0.77 for each ommatidium. Furthermore, it is shown that the transmission of the compound eye can be improved by 2.3% for the wavelength of 632.8 nm and a clearer image can be formed by the fabricated compound eye with antireflective nanostructures compared with that without antireflective nanostructures. In addition, the developed hybrid manufacturing method can be adapted to the fabrication of other complex micro- and nanodevices for photonics or other research areas.

Egomotion estimation with optic flow and air velocity sensors

We develop a method that allows a flyer to estimate its own motion (egomotion), the wind velocity, ground slope, and flight height using only inputs from onboard optic flow and air velocity sensors. Our artificial algorithm demonstrates how it could be possible for flying insects to determine their absolute egomotion using their available sensors, namely their eyes and wind sensitive hairs and antennae. Although many behaviors can be performed by only knowing the direction of travel, behavioral experiments indicate that odor tracking insects are able to estimate the wind direction and control their absolute egomotion (i.e., groundspeed). The egomotion estimation method that we have developed, which we call the opto-aeronautic algorithm, is tested in a variety of wind and ground slope conditions using a video recorded flight of a moth tracking a pheromone plume. Over all test cases that we examined, the algorithm achieved a mean absolute error in height of 7% or less. Furthermore, our algorithm is suitable for the navigation of aerial vehicles in environments where signals from the Global Positioning System are unavailable.

Odometry and insect navigation

Animals have needed to find their way about almost since a free-living life style evolved. Particularly, if an animal has a home – shelter or nesting site – true navigation becomes necessary to shuttle between this home and areas of other activities, such as feeding. As old as navigation is in the animal kingdom, as diverse are its mechanisms and implementations, depending on an organism's ecology and its endowment with sensors and actuators. The use of landmarks for piloting or the use of trail pheromones for route following have been examined in great detail and in a variety of animal species. The same is true for senses of direction – the compasses for navigation – and the construction of vectors for navigation from compass and distance cues. The measurement of distance itself – odometry – has received much less attention. The present review addresses some recent progress in the understanding of odometers in invertebrates, after outlining general principles of navigation to put odometry in its proper context. Finally, a number of refinements that increase navigation accuracy and safety are addressed.

Neural action fields for optic flow based navigation: a simulation study of the fly lobula plate network

Optic flow based navigation is a fundamental way of visual course control described in many different species including man. In the fly, an essential part of optic flow analysis is performed in the lobula plate, a retinotopic map of motion in the environment. There, the so-called lobula plate tangential cells possess large receptive fields with different preferred directions in different parts of the visual field. Previous studies demonstrated an extensive connectivity between different tangential cells, providing, in principle, the structural basis for their large and complex receptive fields. We present a network simulation of the tangential cells, comprising most of the neurons studied so far (22 on each hemisphere) with all the known connectivity between them. On their dendrite, model neurons receive input from a retinotopic array of Reichardt-type motion detectors. Model neurons exhibit receptive fields much like their natural counterparts, demonstrating that the connectivity between the lobula plate tangential cells indeed can account for their complex receptive field structure. We describe the tuning of a model neuron to particular types of ego-motion (rotation as well as translation around/along a given body axis) by its ‘action field’. As we show for model neurons of the vertical system (VS-cells), each of them displays a different type of action field, i.e., responds maximally when the fly is rotating around a particular body axis. However, the tuning width of the rotational action fields is relatively broad, comparable to the one with dendritic input only. The additional intra-lobula-plate connectivity mainly reduces their translational action field amplitude, i.e., their sensitivity to translational movements along any body axis of the fly.

Control theoretic interpretation of directional motion preferences in optic flow processing interneurons

In this article, we formalize the processing of optic flow in identified fly lobula plate tangential cells and develop a control theoretic framework that suggests how the signals of these cells may be combined and used to achieve reflex-like navigation behavior. We show that this feedback gain synthesis task can be cast as a combined static state estimation and linear feedback control problem. Our framework allows us to analyze and determine the relationship between optic flow measurements and actuator commands, which greatly simplifies the implementation of biologically inspired control architectures on terrestrial and aerial robotic platforms.

Motion parallax contribution to perception of self-motion and depth

The object of this study is to mathematically specify important characteristics of visual flow during translation of the eye for the perception of depth and self-motion. We address various strategies by which the central nervous system may estimate self-motion and depth from motion parallax, using equations for the visual velocity field generated by translation of the eye through space. Our results focus on information provided by the movement and deformation of three-dimensional objects and on local flow behavior around a fixated point. All of these issues are addressed mathematically in terms of definite equations for the optic flow. This formal characterization of the visual information presented to the observer is then considered in parallel with other sensory cues to self-motion in order to see how these contribute to the effective use of visual motion parallax, and how parallactic flow can, conversely, contribute to the sense of self-motion. This article will focus on a central case, for understanding of motion parallax in spacious real-world environments, of monocular visual cues observable during pure horizontal translation of the eye through a stationary environment. We suggest that the global optokinetic stimulus associated with visual motion parallax must converge in significant fashion with vestibular and proprioceptive pathways that carry signals related to self-motion. Suggestions of experiments to test some of the predictions of this study are made.

Saccadic flight strategy facilitates collision avoidance: closed-loop performance of a cyberfly

Behavioural and electrophysiological experiments suggest that blowflies employ an active saccadic strategy of flight and gaze control to separate the rotational from the translational optic flow components. As a consequence, this allows motion sensitive neurons to encode during translatory intersaccadic phases of locomotion information about the spatial layout of the environment. So far, it has not been clear whether and how a motor controller could decode the responses of these neurons to prevent a blowfly from colliding with obstacles. Here we propose a simple model of the blowfly visual course control system, named cyberfly, and investigate its performance and limitations. The sensory input module of the cyberfly emulates a pair of output neurons subserving the two eyes of the blowfly visual motion pathway. We analyse two sensory-motor interfaces (SMI). An SMI coupling the differential signal of the sensory neurons proportionally to the yaw rotation fails to avoid obstacles. A more plausible SMI is based on a saccadic controller. Even with sideward drift after saccades as is characteristic of real blowflies, the cyberfly is able to successfully avoid collisions with obstacles. The relative distance information contained in the optic flow during translatory movements between saccades is provided to the SMI by the responses of the visual output neurons. An obvious limitation of this simple mechanism is its strong dependence on the textural properties of the environment.

Robust coding of flow-field parameters by axo-axonal gap junctions between fly visual interneurons

Complex flight maneuvers require a sophisticated system to exploit the optic flow resulting from moving images of the environment projected onto the retina. In the fly's visual course control center, the lobula plate, 10 so-called vertical system (VS) cells are thought to match, with their complex receptive fields, the optic flow resulting from rotation around different body axes. However, signals of single VS cells are unreliable indicators of such optic flow parameters in the context of their noisy, texture-dependent input from local motion measurements. Here we propose an alternative encoding scheme based on network simulations of biophysically realistic compartmental models of VS cells. The simulations incorporate recent data about the highly selective connectivity between VS cells consisting of an electrical axo-axonal coupling between adjacent cells and a reciprocal inhibition between the most distant cells. We find that this particular wiring performs a linear interpolation between the output signals of VS cells, leading to a robust representation of the axis of rotation even in the presence of textureless patches of the visual surround.

A `bright zone' in male hoverfly (Eristalis tenax) eyes and associated faster motion detection and increased contrast sensitivity

Eyes of the hoverfly Eristalis tenax are sexually dimorphic such that males have a fronto-dorsal region of large facets. In contrast to other large flies in which large facets are associated with a decreased interommatidial angle to form a dorsal `acute zone' of increased spatial resolution, we show that a dorsal region of large facets in males appears to form a `bright zone' of increased light capture without substantially increased spatial resolution. Theoretically, more light allows for increased performance in tasks such as motion detection. To determine the effect of the bright zone on motion detection, local properties of wide field motion detecting neurons were investigated using localized sinusoidal gratings. The pattern of local preferred directions of one class of these cells, the HS cells, in Eristalis is similar to that reported for the blowfly Calliphora. The bright zone seems to contribute to local contrast sensitivity; high contrast sensitivity exists in portions of the receptive field served by large diameter facet lenses of males and is not observed in females. Finally, temporal frequency tuning is also significantly faster in this frontal portion of the world, particularly in males, where it overcompensates for the higher spatial-frequency tuning and shifts the predicted local velocity optimum to higher speeds. These results indicate that increased retinal illuminance due to the bright zone of males is used to enhance contrast sensitivity and speed motion detector responses. Additionally, local neural properties vary across the visual world in a way not expected if HS cells serve purely as matched filters to measure yaw-induced visual motion.

Local visual homing by matched-filter descent in image distances

In natural images, the distance measure between two images taken at different locations rises smoothly with increasing distance between the locations. This fact can be exploited for local visual homing where the task is to reach a goal location that is characterized by a snapshot image: descending in the image distance will lead the agent to the goal location. To compute an estimate of the spatial gradient in the distance measure, its value must be sampled at three noncollinear points. An animal or robot would have to insert exploratory movements into its home trajectory to collect these samples. Here we suggest a method based on the matched-filter concept that allows one to estimate the gradient without exploratory movements. Two matched filters--optical flow fields resulting from translatory movements in the horizontal plane--are used to predict two images in perpendicular directions from the current location. We investigate the relation to differential flow methods applied to the local homing problem and show that the matched-filter approach produces reliable homing behavior on image databases. Two alternative methods that only require a single matched filter are suggested. The matched-filter concept is also applied to derive a home-vector equation for a Fourier-based parameter method.

Movement-induced motion signal distributions in outdoor scenes

The movement of an observer generates a characteristic field of velocity vectors on the retina (Gibson 1950). Because such optic flow-fields are useful for navigation, many theoretical, psychophysical and physiological studies have addressed the question how ego-motion parameters such as direction of heading can be estimated from optic flow. Little is known, however, about the structure of optic flow under natural conditions. To address this issue, we recorded sequences of panoramic images along accurately defined paths in a variety of outdoor locations and used these sequences as input to a two-dimensional array of correlation-based motion detectors (2DMD). We find that (a) motion signal distributions are sparse and noisy with respect to local motion directions; (b) motion signal distributions contain patches (motion streaks) which are systematically oriented along the principal flow-field directions; (c) motion signal distributions show a distinct, dorso-ventral topography, reflecting the distance anisotropy of terrestrial environments; (d) the spatiotemporal tuning of the local motion detector we used has little influence on the structure of motion signal distributions, at least for the range of conditions we tested; and (e) environmental motion is locally noisy throughout the visual field, with little spatial or temporal correlation; it can therefore be removed by temporal averaging and is largely over-ridden by image motion caused by observer movement. Our results suggest that spatial or temporal integration is important to retrieve reliable information on the local direction and size of motion vectors, because the structure of optic flow is clearly detectable in the temporal average of motion signal distributions. Ego-motion parameters can be reliably retrieved from such averaged distributions under a range of environmental conditions. These observations raise a number of questions about the role of specific environmental and computational constraints in the processing of natural optic flow.