Search

Australia is one of the most culturally diverse countries in the world. Since World War II, seven million immigrants from more than 150 countries have settled in Australia. Since that time, Federal governmental changes to its policies on immigration has recognized the importance of cultural diversity in its population. Educational institutions have also responded by initiating a variety of strategies and developing curricula aimed at achieving equitable education and social outcomes and promote the acceptance of people from ethnically diverse backgrounds. This paper first examines how Australia has developed policy that has enabled education to become its third largest export market worth more than AUD$18 billion in 2014-2015. It focuses on some of the state and federal government policies that have encouraged the internationalization within the primary, secondary and tertiary education sectors and then gives an overview of some of the institutional strategies and policies that have been implemented at the Royal Melbourne Institute of Technology (RMIT University) at university level and geospatial science undergraduate discipline level. A range of challenges at the governmental, institutional and discipline level faced by those individual academics wishing to incorporate internationalization into their respective curricula are investigated.

EM27/SUN devices are portable solar-viewing Fourier transform spectrometers (FTSs) that are being widely used to constrain measurements of greenhouse gas emissions and validate satellite trace gas measurements. On a 6-week-long campaign in the summer of 2018, four EM27/SUN devices were taken to five Total Carbon Column Observing Network (TCCON) stations in North America, to measure side by side, to better understand their durability, the accuracy and precision of retrievals from their trace gas measurements, and to constrain site-to-site bias among TCCON sites. We developed new EM27/SUN data products using both previous and current versions of the retrieval algorithm (GGG2014 and GGG2020) and used coincident AirCore measurements to tie the gas retrievals to the World Meteorological Organization (WMO) trace gas standard scales. We also derived air-mass-dependent correction factors for the EM27/SUN devices. Pairs of column-averaged dry-air mole fractions (denoted with an X) measured by the EM27/SUN devices remained consistent compared to each other during the entire campaign, with a 10 min averaged precision of 0.3 ppm (parts per million) for XCO 2 , 1.7 ppb (parts per billion) for XCH 4 , and 2.5 ppb for XCO. The maximum biases between TCCON stations were reduced in GGG2020 relative to GGG2014 from 1.3 to 0.5 ppm for XCO 2 and from 5.4 to 4.3 ppb for XCH 4 but increased for XCO from 2.2 to 6.1 ppb. The increased XCO biases in GGG2020 are driven by measurements at sites influenced by urban emissions (Caltech and the Armstrong Flight Research Center) where the priors overestimate surface CO. In addition, in 2020, one EM27/SUN instrument was sent to the Canadian Arctic TCCON station at Eureka, and side-by-side measurements were performed in March–July. In contrast to the other TCCON stations that showed an improvement in the biases with the newer version of GGG, the biases between Eureka's TCCON measurements and those from the EM27/SUN degraded with GGG2020, but this degradation was found to be caused by a temperature dependence in the EM27/SUN oxygen retrievals that is not apparent in the GGG2014 retrievals. [ABSTRACT FROM AUTHOR]

Access to healthcare, including physiotherapy, is increasingly occurring through virtual formats. At-home adherence to physical therapy programs is often poor and few tools exist to objectively measure participation. The aim of this study was to develop and evaluate the potential for performing automatic, unsupervised video-based monitoring of at-home low-back and shoulder physiotherapy exercises using a mobile phone camera. Joint locations were extracted from the videos of healthy subjects performing low-back and shoulder physiotherapy exercises using an open source pose detection framework. A convolutional neural network was trained to classify physiotherapy exercises based on the segments of keypoint time series data. The model's performance as a function of input keypoint combinations was studied in addition to its robustness to variation in the camera angle. The CNN model achieved optimal performance using a total of 12 pose estimation landmarks from the upper and lower body (low-back exercise classification: 0.995 ± 0.009; shoulder exercise classification: 0.963 ± 0.020). Training the CNN on a variety of angles was found to be effective in making the model robust to variations in video filming angle. This study demonstrates the feasibility of using a smartphone camera and a supervised machine learning model to effectively classify at-home physiotherapy participation and could provide a low-cost, scalable method for tracking adherence to physical therapy exercise programs in a variety of settings. [ABSTRACT FROM AUTHOR]

EM27/SUNs are portable solar-viewing Fourier Transform Spectrometers (FTSs) that are being widely used to constrain measurements of greenhouse gas emissions and validate satellite trace gas measurements. On a six-week-long campaign in the summer of 2018, four EM27/SUNs were taken to five Total Carbon Column Observing Network (TCCON) stations in North America to measure side-by-side to better understand their durability, as well as the accuracy and precision of retrievals from their trace gas measurements and to constrain site-to-site bias among TCCON sites. We developed new EM27/SUN data products using both previous and current versions of the retrieval algorithm (GGG2014 and GGG2020) and used coincident AirCore measurements to tie the gas retrievals to the World Meteorological Organization (WMO) trace gas standard scales. We also derived airmass-dependent correction factors for the EM27/SUNs. Pairs of column-averaged dry-air mole fractions (denoted with an X) measured by the EM27/SUNs remained consistent compared to each other during the entire campaign, with a 10-minute averaged precision of 0.3 ppm for XCO2, 1.7 ppb for XCH4 and 2.5 ppb for XCO. The maximum biases between TCCON stations were reduced in GGG2020 relative to GGG2014 from 1.3 ppm to 0.5 ppm for XCO2 and from 5.4 ppb to 4.3 for XCH4 but increased for XCO from 2.2 to 6.1 ppb. The increased XCO biases in GGG2020 are driven by measurements at sites influenced by urban emissions (Caltech and AFRC) where the priors overestimate surface CO. In addition in 2020, one EM27/SUN instrument was sent to the Canadian Arctic TCCON station at Eureka and side-by-side measurements were performed in March–July. In contrast to the other TCCON stations that showed an improvement in the biases with the newer version of GGG, the biases between Eureka's TCCON measurements and those from the EM27/SUN degraded with GGG2020, but this degradation was found to be caused by a temperature dependence in the EM27/SUN oxygen retrievals that is not apparent in the GGG2014 retrievals. [ABSTRACT FROM AUTHOR]

A significant challenge for a supervised learning approach to inertial human activity recognition is the heterogeneity of data generated by individual users, resulting in very poor performance for some subjects. We present an approach to personalized activity recognition based on deep feature representation derived from a convolutional neural network (CNN). We experiment with both categorical cross-entropy loss and triplet loss for training, and describe a novel loss function based on subject triplets. We evaluate these methods on three publicly available inertial human activity recognition datasets (MHEALTH, WISDM, and SPAR) comparing classification accuracy, out-of-distribution activity detection, and generalization to new activity classes. The proposed triplet algorithm achieved an average 96.7% classification accuracy across tested datasets versus the 87.5% achieved by the baseline CNN algorithm. We demonstrate that personalized algorithms, and, in particular, the proposed novel triplet loss algorithms, are more robust to inter-subject variability and thus exhibit better performance on classification and out-of-distribution detection tasks. [ABSTRACT FROM AUTHOR]

Background: Physiotherapy is a critical element in the successful conservative management of low back pain (LBP). A gold standard for quantitatively measuring physiotherapy participation is crucial to understanding physiotherapy adherence in managing recovery from LBP. Objective: This study aimed to develop and evaluate a system with wearable inertial sensors to objectively detect the performance of unsupervised exercises for LBP comprising movement in multiple planes and sitting postures. Methods: A quantitative classification design was used within a machine learning framework to detect exercise performance and posture in a cohort of healthy participants. A set of 8 inertial sensors were placed on the participants, and data were acquired as they performed 7 McKenzie low back exercises and 3 sitting posture positions. Engineered time series features were extracted from the data and used to train 9 models by using a 6-fold cross-validation approach, from which the best 2 models were selected for further study. In addition, a convolutional neural network was trained directly on the time series data. A feature importance analysis was performed to identify sensor locations and channels that contributed the most to the models. Finally, a subset of sensor locations and channels was included in a hyperparameter grid search to identify the optimal sensor configuration and best performing algorithms for exercise and posture classification. The final models were evaluated using the F1 score in a 10-fold cross-validation approach. Results: In total, 19 healthy adults with no history of LBP each completed at least one full session of exercises and postures. Random forest and XGBoost (extreme gradient boosting) models performed the best out of the initial set of 9 engineered feature models. The optimal hardware configuration was identified as a 3-sensor setup--lower back, left thigh, and right ankle sensors with acceleration, gyroscope, and magnetometer channels. The XGBoost model achieved the highest exercise (F1 score: mean 0.94, SD 0.03) and posture (F1 score: mean 0.90, SD 0.11) classification scores. The convolutional neural network achieved similar results with the same sensor locations, using only the accelerometer and gyroscope channels for exercise classification (F1 score: mean 0.94, SD 0.02) and the accelerometer channel alone for posture classification (F1 score: mean 0.88, SD 0.07). Conclusions: This study demonstrates the potential of a 3-sensor lower body wearable solution (eg, smart pants) that can identify exercises in multiple planes and proper sitting postures, which is suitable for the treatment of LBP. This technology has the potential to improve the effectiveness of LBP rehabilitation by facilitating quantitative feedback, early problem diagnosis, and possible remote monitoring. [ABSTRACT FROM AUTHOR]

The Gunbower Yemurriki Map details important information about Barapa Country and Barapa people for the purpose of education. The Barapa are the Traditional Owners of the lands north and south of the Murray River around Cohuna, Australia and are working with natural resource agencies to identify and map cultural assets on traditional lands, particularly in relation to water resources. The Gunbower Yemurriki Map has been developed through participatory cultural mapping processes to demonstrate the community connection to water and the wider cultural landscape. Yemurriki is the Barapa word for Country. The map developed and presented in this study will be used to educate the local non-indigenous community about Barapa cultural values and to act as a teaching aid for younger Barapa people. The map depicts stories, totems, and places identified within the landscape. All the information included is what the Barapa consider public and educational. [ABSTRACT FROM AUTHOR]

Land use changes associated with urbanization often have negative impacts on scenic beauty. In this paper, we explore and compare the visual impact of two different scenarios of urban growth including historical urban growth (HUG) and aesthetically sound urban growth (AUG) with two different categories of height for buildings in the city of Gorgan, Iran. This was done by viewshed and landscape metric analysis of different viewpoints and 3D representation of each scenario. The results show that with the AUG scenario, viewsheds are less impacted by new developments than the HUG scenario in all the viewpoints. It can be concluded that building locations can considerably affect the landscape visibility while building height does not impact to the same degree as location. The results of this research, as a Spatial Decision Support System, would help the managers for better understanding of different patterns of urbanization and its effect on landscape view. [ABSTRACT FROM AUTHOR]

This study demonstrates the integration of landscape aesthetic quality and probable urban growth patterns in urban landscape modelling. This was performed using SLEUTH as a scenario-based urban growth model in Gorgan City of Iran. Future urbanization was predicted under developing three different scenarios including historical, managed and aesthetically sound urban growth up to the year 2030. Multi-Layer Perceptron neural network model was conducted for mapping the aesthetic suitability of the study area. The aesthetic suitability layer was used in the third scenario of SLEUTH model as the excluded layer to protect the scenic patches in future. The results showed that by correct implementation of urban growth policies, 323 ha in the second scenario and 650 ha in the third scenario would be saved. This integrated model would help the planners for a better management of urban landscapes as a Spatial Decision Support System. [ABSTRACT FROM AUTHOR]

Stresses acting on fault systems before major earthquakes can produce thermal anomalies; these abnormalities can be observed using multi-sensor satellite data. Moderate resolution spectro-radiometer on board the terra and aqua satellites can provide thermal infrared (TIR) imaging data for land and ocean. These TIR data have recorded short-lived thermal anomaly prior to major earthquakes. It is suggested by others that these electromagnetic (EM) phenomena relate to stress build up before earthquakes. The objective of this study is to find an association between spatial extent and temporal evolution of thermal anomalies and known major earthquakes near the boundary of Nazca plate and South American plate. Our approach is to map the TIR transient fields from polar orbiting satellites and analysing those data using time series temperature plots to detect the abnormal thermal trends before the earthquake. This study concentrated on marine earthquakes to detect the changes in both land and ocean before seismic activity. [ABSTRACT FROM AUTHOR]

Australia is one of the most culturally diverse countries in the world. Since World War II, seven million immigrants from more than 150 countries have settled in Australia. Since that time, Federal governmental changes to its policies on immigration has recognized the importance of cultural diversity in its population. Educational institutions have also responded by initiating a variety of strategies and developing curricula aimed at achieving equitable education and social outcomes and promote the acceptance of people from ethnically diverse backgrounds. This paper first examines how Australia has developed policy that has enabled education to become its third largest export market worth more than AUD$18 billion in 2014–2015. It focuses on some of the state and federal government policies that have encouraged the internationalization within the primary, secondary and tertiary education sectors and then gives an overview of some of the institutional strategies and policies that have been implemented at the Royal Melbourne Institute of Technology (RMIT University) at university level and geospatial science undergraduate discipline level. A range of challenges at the governmental, institutional and discipline level faced by those individual academics wishing to incorporate internationalization into their respective curricula are investigated. [ABSTRACT FROM AUTHOR]

Geographic clustering of industries for mutual economic benefit has long been recognised. The concept of ‘externalities’ introduced by Alfred Marshall early last century rely on agglomeration of specialised industry within a geographic area. However, only recently has cluster modelling been applied to the tourism and hospitality (T&H) industry. The aim of this paper, therefore, is twofold: first is to develop a cluster-based theoretical framework for delineating geographic boundaries of T&H clusters, and second is to identify the underlying factors that drive their form and shape. Drawing on employment data as the basis for co-location of T&H firms, spatial econometrics techniques are applied to model the spatial clustering of T&H employment in Victoria, Australia. Results show that rural tourism regions have higher levels of employment in tourism operational services whereas employment in city-based regions is more concentrated in hospitality services. Our findings, when normalised as a percentage of total employment, show that rural and regional Victoria ranks most highly as employers in the T&H industry. Adopting a range of spatial metrics, we show that T&H clustering throughout Victoria is largely driven by six location-specific factors: (1) the availability of tourism attractions; (2) proximity to the coast; (3) the road density network; (4) accessibility to employment within the Melbourne CBD; (5) the scale of the regional economy; and, (6) the advantages and disadvantages associated with economic resources. We conclude that the cluster-led strategy pose a number of challenges for tourism planners to promote regional tourism. Nevertheless, results from this study indicate that T&H employment clustering creates a more cohesive spatial structure that could support economic development and better connectedness of tourism destinations. These clusters could act as service hubs to their wider catchment areas where visitors are encouraged to stay overnight and travel to specific sights during the day. [ABSTRACT FROM PUBLISHER]

As the study of cinema has increasingly turned to the examination of economic ebbs and industrial flows, rather than focussing its attention solely on the critical evaluation of the films themselves, new analytic techniques and tools have been adopted (and adapted) by film scholars. Key amongst these is the use of innovative visualization techniques that can assist in the understanding of the spatial and temporal features of film industry practices. However, like the cinema itself, visualization carries its own spatial and temporal dimension. This article explores some of the benefits and limitations that derive from the use of spatial visualization technologies in the field of cinema studies. In particular, this research presents a new holistic multivariate approach to spatio-temporal visualization for point based historical data. This method has been developed through extending the spatial presence in timeline graphics and through meaningful spatial classification and representation. [ABSTRACT FROM AUTHOR]

This paper presents a study which explored a range of techniques for analysing the spatial relationship between population density and geographical features of interest at a global scale. Three categories of spatial analysis techniques were explored: traditional methods, spatial autocorrelation statistics and regression analysis. The correlation between the spatial distribution of Australian cinema screens and a global gridded population density dataset were used as the case study for the analysis. [ABSTRACT FROM AUTHOR]

Cinema data is characteristically complex, heterogeneous and interlinked. Rather than relying on simple information retrieval techniques, researchers are increasingly turning to the creative exploration and reapplication of data in order to more fully explore the meaning of newly available and diverse data sets. In this context, the cinema historian becomes the creator of visual texts which can be assessed for both their interpretive insight and their aesthetic qualities. This paper presents four research projects that use different spatio-temporal visualization techniques to understand the industrial dynamics of post-war film exhibition and distribution in Australia. The research integrates work by a group of inter-disciplinary investigators into the effectiveness of techniques such as dendritic mapping, Circos circular visualizations, animation, cartogram mapping, and multivariate visualization for the study of cinema circuits and operations at a number of scales. [ABSTRACT FROM AUTHOR]

The nature and mission of universities have changed over the past two decades. The move towards mass education with decreased levels of state support has come with greater levels of accountability to stakeholders, including students, employers and the state itself. Graduates are expected to exhibit greater degrees of employability: that is they are more likely to gain and maintain employment, and to progress in workplaces and build careers. In this paper, we discuss the varying mixes of geographical knowledge, technical competencies and personal attributes that graduates require, before examining the implications on the design of geography curricula and learning practices. [ABSTRACT FROM AUTHOR]

The Operational Linescan System (OLS) onboard the Defense Meteorological Satellite Program (DMSP) group of satellites, unlike other passive remote sensing sensors, is capable of recording the emissions from artificial lights on the earth surface. Along with detecting light from forest fires, shipping fleets and gas flares, the OLS sensor also records the light emitted from cities at night. This paper reports on a study that uses the DMSP Operational Linescan (DMSP-OLS) images with fixed gain settings of 20 dB and 50 dB to model selected metrics used in the Indian census for the state of Maharashtra. The study firstly looks into the utility of non-composited single fixed gain radiance calibrated DMSP-OLS products for proposing a method which might help to build a surrogate method for Indian census. Several parameters are considered in this analysis, with detailed focus on population density, total population and proportion of households with electricity access for 35 districts within the state of Maharashtra. Results show that spatial scale plays an important role in selection of the images and gains. Secondly, this study provides a relative assessment of gain setting for the DMSP-OLS images in an urban Indian context. Images with a gain of 50 dB prove suitable for larger areas while those with a gain of 20 dB give better results at a smaller spatial scale. Statistical analysis and residual maps of spatial distribution of total population and population density validate the result. [ABSTRACT FROM PUBLISHER]

This paper presents a study that investigates the spatial behavioural patterns of international tourists travelling to the City of Melbourne in Australia. It demonstrates that these behavioural patterns are influenced by the differing cultural backgrounds of the visitors. A comparative study based on Hofstede using his four cultural dimensions, power distance, individualism/collectivism, uncertainty avoidance and masculinity/femininity was used to classify these spatial patterns. The study also investigated how the needs of tourists from different counties varied according to their cultural background. The study was based on 278 visitor surveys completed by international visitors to Melbourne city centre attractions. The participants were selected using a random intercept method and were completed at eight key tourist attractions in Melbourne. These surveys sought information with respect to their spatial movements and behaviour, as well as information relating to socio-demographic characteristics for each of the participants. Data were analysed using independent sampled t-tests, one-way analysis of variance (ANOVA) and Pearson correlate analyses. The results showed marked differences between tourists from various national cultures with respect to spatial behavioural patterns in terms of accommodation location, mode of transport to travel, length, direction, type and pattern of movements. This particular study was successful in reinforcing the important role that cultural background plays in destination planning and management. [ABSTRACT FROM AUTHOR]

This article explores temporal variability, referred to as temporal dynamics, in the movement patterns of tourists. Circular statistics were used to compute circular mean times and dispersions for different groups of tourists using data collected with Global Positioning, Systems receivers at Port Campbell national Park in Victoria, Australia over a 2-day period. The results indicate that there is a tendency for tourists to visit the site around noon. The differences in the circular mean times competed for different regions and for different types of tourist are not significant. Nonetheless, kernel density estimation plots enabled variability, exhibited in terms of peak time and mentality in patterns of visitor use, to be detected. [ABSTRACT FROM AUTHOR]

This paper discusses a geographical information system (GIS)-based technique to measure the recreational potential of natural tourist destinations. The technique was developed for a study area in western Victoria, Australia, known as the Grampians National Park (GNP), a popular natural tourist destination. Regression modelling was used to develop a set of predictors of scenic attractiveness derived from data collected via questionnaire administered to a group of university students. The derived output was then overlaid with 'recreational opportunity potential' generated for the region. In the final stage, a spatial model of 'recreational potential' was developed from output maps over the entire geographical area. The study found that the areas with high recreational potential are concentrated around more popular walking tracks in the Wonderland Ranges near Halls Gap in the north-east sector. However, other underdeveloped areas of high recreational potential are identified and suggested as alternative strategic sites to ease potential pressure developing around heavily used walking tracks. Despite the limited sample group, the developed technique offers park managers a method for predicting recreational opportunities. [ABSTRACT FROM AUTHOR]

During every Australian summer fires are common in the south-eastern region of the continent. The combined forces of climate, topography and vegetation make Victoria in particular, one of the most fire prone regions on earth ( DSE 2003 ). Throughout its history, Victoria has seen a number of devastating bushfires, including Black Friday 1939, Ash Wednesday 1983, and more recently in the northeast of the State in 2003. The loss of life combined with the damage caused to land and property results in a heavy cost to the community. In Victoria, two of the organizations involved in fire management are the Victorian Department of Sustainability and Environment (DSE) and the Country Fire Authority (CFA). Both use fire ‘meters’ to determine potential fire behaviour given certain conditions. Values for temperature, wind speed, fuel load and vegetation type are input and a numerical estimate of fire danger given. There are a number of different meters used for different locations and environmental types. The most common meter used in Victoria is the McArthur Meter ( CSIR0 2001b ). The output data from this meter is numerical, and provides no spatial representation of fire danger. This paper looks at a variety of techniques used to visualise the numerical output from the McArthur Forest Fire Danger Meter. The article outlines the different models used by fire managers to simulate a fire situation, to assess future scenarios and for decision making involving fire management. Particular emphasis is placed on the McArthur Forest Fire Danger Meter as this is commonly used by fire departments in Australia. The article then focuses on geographical visualisation and a number of techniques employed to convey spatial information are discussed. The article then goes on to describe the fire simulation prototypes created for a study, a visualisation proof-of-concept product for organizations involved in managing bushfires in Australia. Finally, results from the evaluation of the prototype are presented. [ABSTRACT FROM AUTHOR]

Simple Summary: Recent studies exploring the application of radiomics features in medicine have shown promising results. However, variation in imaging parameters may impact the robustness of these features. Feature robustness may then in turn affect the prediction performance of the machine learning models built upon these features. While numerous studies have tested feature robustness against a variety of imaging parameters, the extent to which feature robustness affects predictions remains unclear. A particularly notable application of radiomics in clinical oncology is the prediction of Human Papillomavirus (HPV) association in Oropharyngeal cancer. In this study we explore how CT scanner type affects the performance of radiomics features for HPV association prediction and highlight the need to implement precautionary approaches so as to minimize this effect. Studies have shown that radiomic features are sensitive to the variability of imaging parameters (e.g., scanner models), and one of the major challenges in these studies lies in improving the robustness of quantitative features against the variations in imaging datasets from multi-center studies. Here, we assess the impact of scanner choice on computed tomography (CT)-derived radiomic features to predict the association of oropharyngeal squamous cell carcinoma with human papillomavirus (HPV). This experiment was performed on CT image datasets acquired from two different scanner manufacturers. We demonstrate strong scanner dependency by developing a machine learning model to classify HPV status from radiological images. These experiments reveal the effect of scanner manufacturer on the robustness of radiomic features, and the extent of this dependency is reflected in the performance of HPV prediction models. The results of this study highlight the importance of implementing an appropriate approach to reducing the impact of imaging parameters on radiomic features and consequently on the machine learning models, without removing features which are deemed non-robust but may contain learning information. [ABSTRACT FROM AUTHOR]

This paper discusses research aimed at identifying key factors influencing the distribution of residential housing renovation employment in metropolitan Melbourne. Using Geographically Weighted Regression (GWR), employment focused on residential housing renovation is modelled using six parameters representing urban space: distance to the central business district, median household income, distance to highways, the number of nearby shopping centres, distance to public open space and accessibility to railway stations. Of the six different explanatory variables, the estimated value of the Ordinary Least Square model for distance to CBD and open space were statistically significant. Mapping the values of local coefficient estimates of independent variables revealed their extent of influence and variation in residential housing renovation employment. [ABSTRACT FROM AUTHOR]

Access to healthcare, including physiotherapy, is increasingly occurring through virtual formats. At-home adherence to physical therapy programs is often poor and few tools exist to objectively measure participation. The aim of this study was to develop and evaluate the potential for performing automatic, unsupervised video-based monitoring of at-home low-back and shoulder physiotherapy exercises using a mobile phone camera. Joint locations were extracted from the videos of healthy subjects performing low-back and shoulder physiotherapy exercises using an open source pose detection framework. A convolutional neural network was trained to classify physiotherapy exercises based on the segments of keypoint time series data. The model's performance as a function of input keypoint combinations was studied in addition to its robustness to variation in the camera angle. The CNN model achieved optimal performance using a total of 12 pose estimation landmarks from the upper and lower body (low-back exercise classification: 0.995 ± 0.009; shoulder exercise classification: 0.963 ± 0.020). Training the CNN on a variety of angles was found to be effective in making the model robust to variations in video filming angle. This study demonstrates the feasibility of using a smartphone camera and a supervised machine learning model to effectively classify at-home physiotherapy participation and could provide a low-cost, scalable method for tracking adherence to physical therapy exercise programs in a variety of settings.