Occupancy Estimation Research Articles

Using static postures to estimate spinal loading during dynamic lifts with participant-specific thoracolumbar musculoskeletal models

Static biomechanical simulations are sometimes used to estimate in vivo kinetic demands because they can be solved efficiently, but this ignores any potential inertial effects. To date, comparisons between static and dynamic analyses of spinal demands have been limited to lumbar joint differences in young males performing sagittal lifts. Here we compare static and dynamic vertebral compressive and shear force estimates during axial, lateral, and sagittal lifting tasks across all thoracic and lumbar vertebrae in older men and women. Participant-specific thoracolumbar full-body musculoskeletal models estimated vertebral forces from recorded kinematics both with and without consideration of dynamic effects, at an identified frame of peak vertebral loading. Static analyses under-predicted dynamic compressive and resultant shear forces, by an average of about 16% for all three lifts across the thoracic and lumbar spine but were highly correlated with dynamic forces (average r2 > .95). The study outcomes have the potential to enable standard clinical and occupational estimates using static analyses.

The development of a convolutional neural network for the automatic detection of Northern Bobwhite Colinus virginianus covey calls

AbstractPassive acoustic monitoring using Autonomous Recording Units (ARUs) is becoming a significant research tool for collecting large amounts of ecological data. Northern bobwhite Colinus virginianus is an economically important game bird whose declining populations are of conservation concern, so efforts to monitor bobwhite abundance using ARUs are being intensified. Yet, manual processing of ARU data is time consuming and often expensive, so developing automatic call detection methods is a key step in acoustic monitoring. We present here the first single species convolutional neural network (CNN) developed purely for automatic bobwhite covey call identification and classification. We demonstrate the value of meaningful data augmentation by including non‐target calls and background noise into our training dataset, as well as evaluating alternative CNN score thresholds and model extrapolation performance. We trained our CNN on 6,682 manually labeled covey calls across three groups of sites within the southeastern USA. Precision and AUC from both CNN classification and individual call detection was high (0.80–0.99), and our model showed strong extrapolation ability across site groups. However, extrapolation performance significantly decreased for sites that were more dissimilar to the training data set if our meaningful data augmentation process was omitted. Our CNN detected significantly more covey calls than manual labeling using Raven Pro software, and processing time was greatly reduced: a single one hour wav file can be now analyzed by the CNN in roughly eight seconds. We also demonstrate using a simple case study that extremely high variability in estimates of bobwhite site occupancy and detection are obtained depending on the method of acoustic data processing (manual versus CNN). Our results suggest that our CNN provides robust and time‐saving analysis of bobwhite covey call acoustic data and can be applied to future research and monitoring projects with high confidence in the performance of the model.

Occupancy estimation in smart buildings using predictive modeling in imbalanced domains

Occupancy estimation in smart buildings using predictive modeling in imbalanced domains

On-Street Parking Occupancy Inference Based on Payment Transactions

A record of spatial and temporal parking occupancy is critical to optimize on-street parking resources and to develop effective parking policies. Such data are often obtained through advanced and costly occupancy monitoring technologies. Moreover, it is usually challenging to integrate bay-level occupancies and parking data from other systems. Accurate occupancy–payment data are required for a wide range of analytical and practical purposes, including but not limited to investigating payment behavior, estimating and forecasting occupancy, and evaluating the efficiency and effectiveness of enforcement policies. This study proposes a metaheuristic optimization algorithm to integrate snapshots of bay-level parking occupancy, captured using simple cameras, with transactions from a conventional parking payment management system. The resulting integrated data were used to develop, calibrate and validate a parking occupancy estimation method utilizing parking payment data only. Details of the design, implementation, and validation of the proposed algorithm and modelling technique are provided. Logistic regression analysis was used to tune parameters of the data integration algorithm. Deep learning, gradient boosting and random forests were used to develop a model of parking occupancy. Evaluation of the algorithm indicated an accuracy of 76% of correct data integration; that is, individual bay occupancies integrated with the correct corresponding payment transactions. The best occupancy estimation model also showed a very high accuracy, with an R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> above 94% and a root mean square error (RMSE) of 1.2 (occupied bays), when tested with a random sample from the integrated data.

A new curb lane monitoring and illegal parking impact estimation approach based on queueing theory and computer vision for cameras with low resolution and low frame rate

The rapid development of the internet of things (IoT), sensing technologies, and machine learning and deep learning techniques, along with the growing variety and volume of data, have yielded new perspectives on how novel technologies can be applied to obtain new sources of curb data to achieve cost-effective curb management. This study presents a new computer vision–based data acquisition and analytics approach for curb lane monitoring and illegal parking impact assessment. The proposed “rank, detect, and quantify impacts” system consists of three main modules: 1) hotspot identification based on rankings generated by local spatial autocorrelation analysis, 2) curb lane occupancy estimation leveraging traffic cameras and computer vision techniques, and 3) illegal parking traffic impact quantification using an M/M/∞ queueing model. To demonstrate the feasibility and validity of the proposed approach, it was tested and empirically validated using field data collected from three case study sites in Midtown Manhattan, New York City (NYC)—one of the most complex urban transportation networks in the world. Different types of curb lane occupancy, including parking and bus lanes, and different frequencies of illegal parking (high, moderate, low) were investigated. The proposed method was proven to be effective even for low resolution and discontinuous video feeds obtained from publicly available traffic cameras. All three case study sites achieved good detection accuracy (86% to 96%) for parking and bus lane occupancy, and acceptable precision and recall in detecting illegal parking events. The queueing model was also proven to effectively quantify link travel time with the appearance of illegal parking events of different frequencies. The proposed “rank, detect, and quantify impacts” system is friendly for large-scale real-time implementation and is highly scalable to help evaluate the impact of other modes such as bike or mobility-on-demand (MOD) services. It can also be easily adopted by other cities to provide transportation agencies with effective data collection and innovative curb space management strategies.

Indoor occupancy estimation for smart utilities: A novel approach based on depth sensors

Occupancy information in indoor spaces is playing an increasingly important role in the development of smart applications. The need for this type of information covers a multitude of domains in the Smart Buildings paradigm such as improving energy saving or occupant comfort. For this reason, we can find many works in the literature focused on occupancy tracking/monitoring using solutions based on RGB cameras and computer vision techniques, sensors and machine learning techniques, or air quality control, among others. But these solutions have limitations. Some of them do not support the tracking of people between spaces, the time to update information is too long, or the system used is too intrusive. This paper presents a solution to estimate the occupancy level in indoor spaces of different areas through depth cameras. This approach also proposes the integration of neural networks to deal with situations where the data collected from the environment is incomplete, filling the gaps caused by occlusion or performance problems. Finally, an occupancy service has been designed and deployed in order to provide occupancy information to other applications, such as evacuation services. The experiments carried out show how it is possible to obtain an accuracy of 90.20% through this approximation. In addition, we face some of the limitations mentioned above: the solution allows tracking movement and occupancy in large spaces without (1) lighting dependencies and (2) the requirement for users to wear devices. This, and the high accuracy obtained make the proposed work a great alternative for occupancy estimation in indoor spaces.

Effectiveness of a commercial lure to attract red fox

In camera trap studies, attractants may be used to increase detection probabilities of wildlife, which may help to improve estimates of abundance and occupancy. Using a semi-experimental approach, we investigated if a commercial, strawberry scented lure increased detection probability and visiting time duration in red fox Vulpes vulpes, and the potential reasons for variation in these parameters. In September 2020, within the Stelvio National Park, central Italian Alps, 32 camera sites were randomly assigned to 4 different treatments: 8 to commercial lure, the target of our investigation; 8 to orange aroma, to test for the “curiosity” effect; 8 to cat kibble, to test for the “likability” effect; 8 to camera trap only, the control test. Detection probability and duration of visiting time were estimated using hurdle negative binomial regression models. Daily detection probability was significantly higher with lure (0.078), orange aroma (0.086), kibble (0.075) than with camera trap only (0.031); in the first day after treatment, the time an animal spent in front of the cameras significantly increased with orange aroma (16.61 s) and kibble (33.78 s) compared to lure (9.97 s) and camera trap only (0.38 s). Our results support the use of lures to improve detection probability and visit duration in red fox, but we could not disentangle the drivers of increased parameter estimates. When consumable costs are considered, the use of the commercial strawberry scented lure does not appear justified for both detection probability and visit duration, and cheaper alternatives may be preferable.

Estimating abundance in unmarked populations of Golden Eagle (Aquila chrysaetos)

Abstract Estimates of species abundance are of key importance in population and ecosystem level research but can be hard to obtain. Study designs using camera traps are increasingly being used for large‐scale monitoring of species that are elusive and/or occur naturally at low densities. Golden eagle (Aquila chrysaetos) is one such species, and we investigate whether existing large‐scale monitoring programs using baited camera traps can be used to estimate the abundance of golden eagles, as an alternative to traditional labour‐intensive searches for active territories and nest sites during the breeding period. The camera‐trap data allowed two measures of abundance to be estimated within each of four main study areas in mid and northern Norway; occupancy was measured as the probability of camera site use, and population size was measured as the number of individuals using camera sites within a study area. Spatial and temporal patterns in occupancy and population size were explored and evaluated against independent estimates of breeding pair density in the study areas. Annual estimates of golden eagle occupancy showed low precision, while estimates of population size were more precise in relation to both estimated and anticipated abundance fluctuations. Estimates of population size may therefore be suitable for monitoring within‐study area temporal abundance trends, while estimates of occupancy seem unsuitable for such for golden eagles. Across study areas, patterns in both average occupancy and average population density estimated from population size were consistent with the spatial pattern in average breeding pair densities (r = 0.99, and r = 0.89, respectively). This suggests that camera‐trap‐based estimates of occupancy and population density, both reflect territory density at large spatial scales. In conclusion, our results suggest that baited camera traps can be a cost‐effective strategy for monitoring the abundance of golden eagles.

Co-occurrence models fail to infer underlying patterns of avoidance and aggregation when closure is violated.

Advances in multi‐species monitoring have prompted an increase in the use of multi‐species occupancy analyses to assess patterns of co‐occurrence among species, even when data were collected at scales likely violating the assumption that sites were closed to changes in the occupancy state for the target species. Violating the closure assumption may lead to erroneous conclusions related to patterns of co‐occurrence among species. Occurrence for two hypothetical species was simulated under patterns of avoidance, aggregation, or independence, when the closure assumption was either met or not. Simulated populations were sampled at two levels (N = 250 or 100 sites) and two scales of temporal resolution for surveys. Sample data were analyzed with conditional two‐species occupancy models, and performance was assessed based on the proportion of simulations recovering the true pattern of co‐occurrence. Estimates of occupancy were unbiased when closure was met, but biased when closure violations occurred; bias increased when sample size was small and encounter histories were collapsed to a large‐scale temporal resolution. When closure was met and patterns of avoidance and aggregation were simulated, conditional two‐species models tended to correctly find support for non‐independence, and estimated species interaction factors (SIF) aligned with predicted values. By contrast, when closure was violated, models tended to incorrectly infer a pattern of independence and power to detect simulated patterns of avoidance or aggregation that decreased with smaller sample size. Results suggest that when the closure assumption is violated, co‐occurrence models often fail to detect underlying patterns of avoidance or aggregation, and incorrectly identify a pattern of independence among species, which could have negative consequences for our understanding of species interactions and conservation efforts. Thus, when closure is violated, inferred patterns of independence from multi‐species occupancy should be interpreted cautiously, and evidence of avoidance or aggregation is likely a conservative estimate of true pattern or interaction.

Behind the Curtain: Learning Occluded Shapes for 3D Object Detection

Advances in LiDAR sensors provide rich 3D data that supports 3D scene understanding. However, due to occlusion and signal miss, LiDAR point clouds are in practice 2.5D as they cover only partial underlying shapes, which poses a fundamental challenge to 3D perception. To tackle the challenge, we present a novel LiDAR-based 3D object detection model, dubbed Behind the Curtain Detector (BtcDet), which learns the object shape priors and estimates the complete object shapes that are partially occluded (curtained) in point clouds. BtcDet first identifies the regions that are affected by occlusion and signal miss. In these regions, our model predicts the probability of occupancy that indicates if a region contains object shapes and integrates this probability map with detection features and generates high-quality 3D proposals. Finally, the occupancy estimation is integrated into the proposal refinement module to generate accurate bounding boxes. Extensive experiments on the KITTI Dataset and the Waymo Open Dataset demonstrate the effectiveness of BtcDet. Particularly for the 3D detection of both cars and cyclists on the KITTI benchmark, BtcDet surpasses all of the published state-of-the-art methods by remarkable margins. Code is released.

Comparison of Two Detection Methods for a Declining Rodent, the Allegheny Woodrat, in Virginia

Abstract Allegheny woodrats Neotoma magister are an imperiled small mammal species most associated with emergent rock habitats in the central Appalachian Mountains and the Ohio River Valley. The monitoring of populations and their spatiotemporal distributions typically has relied on labor-intensive livetrapping. The use of remote-detecting cameras holds promise for being an equally or more effective method to determine species presence, although trap-based captures permit the estimation of other parameters (e.g., survival, population size, site fidelity). In 2017, 2018, and 2020 we compared standard livetrapping with paired cameras for determining site occupancy of Allegheny woodrats in the central Appalachian Mountains of western Virginia. We further examined the influence of baited vs. unbaited cameras at several sites of confirmed occupancy in 2019. We observed that the detection probability using cameras was approximately 1.7 times that of live traps. Also, detection probability at baited camera traps was 1.3–2.0 times that of unbaited camera traps. Estimates of occupancy ranged from 0.44 to 0.49. Our findings suggest that the use of baited remote-detecting cameras provides a more effective method than livetrapping for detecting Allegheny woodrats. Our study provides a framework for the development of a large-scale, long-term monitoring protocol of Allegheny woodrats with the goals of identifying changes in the distribution of the species and quantifying local extinction and colonization rates at emergent rock outcrops and caves throughout the species' known distribution.

Effects of occupancy estimation on abundance-occupancy relationships.

Abundance-occupancy relationships predict that species that occupy more sites are also more locally abundant, where occupancy is usually estimated following the assumption that species can occupy all sampled sites. Here we use the National Ecological Observatory Network small-mammal data to assess whether this assumption affects abundance-occupancy relationships. We estimated occupancy considering all sampled sites (traditional occupancy) and only the sites found within the species geographic range (spatial occupancy) and realized environmental niche (environmental occupancy). We found that when occupancy was estimated considering only sites possible for the species to colonize (spatial and environmental occupancy) weaker abundance-occupancy relationships were observed. This shows that the assumption that the species can occupy all sampled sites directly affects the assessment of abundance-occupancy relationships. Estimating occupancy considering only sites that are possible for the species to colonize will consequently lead to a more robust assessment of abundance-occupancy relationships.

Estimating occupancy from autonomous recording unit data in the presence of misclassifications and detection heterogeneity

Abstract Autonomous recording units (ARUs) are now widely used to survey communities of species. These surveys generate spatially and temporally replicated counts of unmarked animals, but such data typically include false negatives and misclassified detections, both of which may vary across sites in proportion to abundance. These data challenges can bias estimates of occupancy, and the typical approach of verifying individual detections is expensive. We developed a Bayesian implementation of a two‐species, false‐positive N‐mixture model for estimating occupancy from ARU data or other counts of unmarked animals that does not require manual verification. The model accounts for species misclassification and abundance‐induced detection heterogeneity, as well as false negatives. To evaluate this model, we simulated 200 datasets for each of 29 scenarios, including scenarios in which misclassifications outnumbered correct classifications for rare species. We also applied the model to acoustic surveys of bats conducted on Fort Carson Army Post and Piñon Canyon Maneuver Site, Colorado, USA. In the simulation study, bias, coverage and root mean square error for occupancy estimates obtained from the two‐species false‐positive N‐mixture model were superior to metrics obtained from two competing two‐species false‐positive occupancy models. Across 29 scenarios, absolute bias was consistently low (range: −0.03 to 0.07), while coverage averaged 93% (range: 74%–98%). For alternative occupancy models, absolute bias was often high (range: −0.36 to 0.39), and coverage averaged from 47% to 65%. Although our model included an abundance parameter, abundance estimates were not reliable. For two species of Myotis bats, we estimated that 1%–5% of field‐recorded detections were misclassified. Estimated occupancy (0.91 and 0.76) was lower than naïve estimates (1.00 and 0.94). Competing occupancy models implausibly estimated local occupancy of 0.00 at sites with numerous detections. Our two‐species, false‐positive N‐mixture model is significant because it accounts for detection heterogeneity and improves occupancy estimates without expensive manual verification of detections. Our field application indicated that misclassifications were not common, yet affected occupancy inferences. Given that ARUs are increasingly used to survey a broad range of taxa, such an occupancy model could be widely useful.

Comparison of WLAN Probe and Light Sensor-Based Estimators of Bus Occupancy Using Live Deployment Data.

Bus company operators are interested in obtaining knowledge about the number of passengers on their buses—preferably doing so at low deployment costs and in an automated manner, while keeping accuracy high. One solution, widely used in practice, involves deploying a light sensor-based system, counting the people entering and leaving the bus. The light sensor system is simple, but errors accumulate over time, because it is not capable of error correcting. For this reason, the light sensor-based system is compared to a WLAN probe-based system, which has entirely different characteristics. Inaccuracy with the WLAN estimator comes from a need to filter out mobile devices outside the bus and to map the number of detected devices to a number of people. The comparison is performed based on data collected from a real-life deployment in a medium sized German city. The comparison shows the trade-off in selecting either of the two methods. Furthermore, a novel approach for fusion of the light sensor and WLAN estimators is proposed which has a big potential in improving accuracy of both estimators. A fusion approach is proposed that utilizes the different error characteristics for error compensation by calculating compensation terms. The knowledge of Ground Truth is not required as part of this fusion approach for calibration; results show that the approach can find the optimal parameter settings and that it makes this occupancy estimation approach scalable and automated.

Ignoring species availability biases occupancy estimates in single‐scale occupancy models

Abstract Most applications of single‐scale occupancy models do not differentiate between availability and detectability, even though species availability is rarely equal to one. Species availability can be estimated using multi‐scale occupancy models; however, for the practical application of multi‐scale occupancy models, it can be unclear what a robust sampling design looks like and what the statistical properties of the multi‐scale and single‐scale occupancy models are when availability is less than one. Using simulations, we explore the following common questions asked by ecologists during the design phase of a field study: (Q1) what is a robust sampling design for the multi‐scale occupancy model when there are a priori expectations of parameter estimates? (Q2) what is a robust sampling design when we have no expectations of parameter estimates? and (Q3) can a single‐scale occupancy model with a random effects term adequately absorb the extra heterogeneity produced when availability is less than one and provide reliable estimates of occupancy probability? Our results show that there is a tradeoff between the number of sites and surveys needed to achieve a specified level of acceptable error for occupancy estimates using the multi‐scale occupancy model. We also document that when species availability is low (&lt;0.40 on the probability scale), then single‐scale occupancy models underestimate occupancy by as much as 0.40 on the probability scale, produce overly precise estimates, and provide poor parameter coverage. This pattern was observed when a random effects term was and was not included in the single‐scale occupancy model, suggesting that adding a random‐effects term does not adequately absorb the extra heterogeneity produced by the availability process. In contrast, when species availability was high (&gt;0.60), single‐scale occupancy models performed similarly to the multi‐scale occupancy model. Users can further explore our results and sampling designs across a number of different scenarios using the RShiny app https://gdirenzo.shinyapps.io/multi‐scale‐occ/. Our results suggest that unaccounted for availability can lead to underestimating species distributions when using single‐scale occupancy models, which can have large implications on inference and prediction, especially for those working in the fields of invasion ecology, disease emergence, and species conservation.

Multiyear monitoring of threatened iconic arboreal mammals in a mid-elevation conservation reserve in eastern Australia.

Multiyear investigations of population dynamics are fundamental to threatened species conservation. We used multiseason occupancy based on spotlight surveys to investigate dynamic occupancy of the koala and the greater glider over an 8‐year period that encompassed a severe drought in year 6. We combined our occupancy estimates with literature estimates of density to estimate the population sizes of these species within the focal conservation reserve. Both species showed substantial yearly variation in the probability of detection (koala: 0.13–0.24; greater glider: 0.12–0.36). Detection of the koala did not follow any obvious pattern. Low detection of the greater glider coincided with the drought and two subsequent years. We suggest the low detection reflected a decline in abundance. The probability of occupancy of the koala was estimated to be 0.88 (95% CI: 0.75–1.0) in year 8. Autonomous recording units were also used in year 8, enabling an independent occupancy estimate of 0.80 (0.64–0.90). We found no evidence of a drought‐induced decline in the koala. Habitat variables had a weak influence on koala occupancy probabilities. The probability of occupancy of the greater glider changed little over time, from 0.52 (95% CI: 0.24–0.81) to 0.63 (0.42–0.85) in year 8. Modeling suggested that the probability of colonization was positively influenced by the percentage cover of rainforest. Increased cover of these nonbrowse trees may reflect thermal buffering, site productivity, or soil moisture. We estimate that our study reserve is likely to contain >900 adult koalas and >2400 adult greater gliders. These are among some of the first reserve‐wide estimates for these species. Our study reserve can play an important role in the conservation of both species.

Automated bird sound classifications of long-duration recordings produce occupancy model outputs similar to manually annotated data

Abstract Occupancy modeling is used to evaluate avian distributions and habitat associations, yet it typically requires extensive survey effort because a minimum of 3 repeat samples are required for accurate parameter estimation. Autonomous recording units (ARUs) can reduce the need for surveyors on-site, yet their utility was limited by hardware costs and the time required to manually annotate recordings. Software that identifies bird vocalizations may reduce the expert time needed if classification is sufficiently accurate. We assessed the performance of BirdNET—an automated classifier capable of identifying vocalizations from &amp;gt;900 North American and European bird species—by comparing automated to manual annotations of recordings of 13 breeding bird species collected in northwestern California. We compared the parameter estimates of occupancy models evaluating habitat associations supplied with manually annotated data (9-min recording segments) to output from models supplied with BirdNET detections. We used 3 sets of BirdNET output to evaluate the duration of automatic annotation needed to approach manually annotated model parameter estimates: 9-min, 87-min, and 87-min of high-confidence detections. We incorporated 100 3-s manually validated BirdNET detections per species to estimate true and false positive rates within an occupancy model. BirdNET correctly identified 90% and 65% of the bird species a human detected when data were restricted to detections exceeding a low or high confidence score threshold, respectively. Occupancy estimates, including habitat associations, were similar regardless of method. Precision (proportion of true positives to all detections) was &amp;gt;0.70 for 9 of 13 species, and a low of 0.29. However, processing of longer recordings was needed to rival manually annotated data. We conclude that BirdNET is suitable for annotating multispecies recordings for occupancy modeling when extended recording durations are used. Together, ARUs and BirdNET may benefit monitoring and, ultimately, conservation of bird populations by greatly increasing monitoring opportunities.

Weakly Supervised Occupancy Prediction Using Training Data Collected via Interactive Learning.

Accurate and timely occupancy prediction has the potential to improve the efficiency of energy management systems in smart buildings. Occupancy prediction heavily depends on historical occupancy-related data collected from various sensor sources. Unfortunately, a major problem in that context is the difficulty to collect training data. This situation inspired us to rethink the occupancy prediction problem, proposing the use of an original principled approach based on occupancy estimation via interactive learning to collect the needed training data. Following that, the collected data, along with various features, were fed into several algorithms to predict future occupancy. This paper mainly proposes a weakly supervised occupancy prediction framework based on office sensor readings and occupancy estimations derived from an interactive learning approach. Two studies are the main emphasis of this paper. The first is the prediction of three occupancy states, referred to as discrete states: absence, presence of one occupant, and presence of more than one occupant. The purpose of the second study is to anticipate the future number of occupants, i.e., continuous states. Extensive simulations were run to demonstrate the merits of the proposed prediction framework’s performance and to validate the interactive learning-based approach’s ability to contribute to the achievement of effective occupancy prediction. The results reveal that LightGBM, a machine learning model, is a better fit for short-term predictions than known recursive neural networks when dealing with a limited dataset. For a 24 h window forecast, LightGBM improved accuracy from 38% to 50%, which is an excellent result for non-aggregated data (single office).

Carnivore Occupancy within the Early Successional Habitat of New England Cottontails

Abstract Predation pressure from carnivores can shape ecological communities and has significant consequences for prey species that are declining or recovering from historical declines. New England cottontails Sylvilagus transitionalis are a species of greatest conservation need in Connecticut and are experiencing continued decline associated with habitat loss. Restoration of early successional habitat is underway to address the most significant threat to their populations. However, one of the largest documented sources of mortality is associated with several key predators and remains a threat to recovery efforts. We sought to develop species-specific occupancy estimates of carnivores in early successional habitat and relate our findings to the potential recovery of New England cottontails. We conducted camera surveys at 34 sites in early successional habitat in or near New England cottontail focus areas throughout Connecticut and used the program MARK to estimate occupancy and detectability from detection data. We found key predators in early successional habitat, but their detectability was generally low. Occupancy was highest for coyotes Canis latrans, and regional occupancy differed only for bobcats Lynx rufus. Covariates that influenced parameter estimates in our models included high road densities and the intensity of New England cottontail and eastern cottontail Sylvilagus floridanus detections. Expanding carnivores, particularly coyote and bobcat, may place additional pressure on New England cottontail recovery in the state, but restoration efforts that promote contiguous habitat and reduce isolated patches where predation risk is higher will improve their chances of a long-term recovery.

Transfer learning for estimating occupancy and recognizing activities in smart buildings

Activities Recognition (AR) and Occupancy Estimation (OE) are vital to many smart systems that work on providing good services in smart buildings. Many applications, such as energy management need information like activities and occupancy to provide good assistance. Most of the previous research about AR and OE focused on applying supervised machine learning methods. Researchers train a model and evaluate it using data collected from the same environment (Domain). A model trained in a specific domain will not generalize well in other domains. Creating a trained model to every environment is not feasible due to the lack of data. Collecting sufficient data can be time consuming and infeasible in some cases. Computational power can be a challenge for researchers by increasing the training time due to the lack of the required computing resources. Using traditional machine learning methods, the obtained performance may be unsatisfactory, and can not lead to optimal solutions. For all these reasons, we need a solution that helps us overcome the stated problem and obtain models with acceptable results. In this work, we present and discuss different transfer learning methods that help us transfer knowledge from a source domain to a target domain. The goal is to reuse as much as possible information from the source domain to enhance the performance of the model at the target domain. This type of approaches will solve the problems mentioned before such as the lack of data and will provide us with good results due to the use of knowledge from multiple source domains. We tested five Transfer learning (TL) approaches: a principal component analysis (PCA)-like method that creates a transformation like the PCA transformation and apply it to the data to create new common domain, a PCA based method that creates common domain using PCA, a PCA-SMOTE method that balances the data and creates common domain, a basic method based on a simple matching between similar features from source and target domain, and a sparse coding-based method that creates a common domain where the data representation will be as sparse as possible. The impressive results that we obtained in both tasks prove that the presented methods can be applied to transfer knowledge across different domains.