Rita Brugarolas

Behavior recognition based on machine learning algorithms for a wireless canine machine interface

Training and handling working dogs is a costly process and requires specialized skills and techniques. Less subjective and lower-cost training techniques would not only improve our partnership with these dogs but also enable us to benefit from their skills more efficiently. To facilitate this, we are developing a canine body-area-network (cBAN) to combine sensing technologies and computational modeling to provide handlers with a more accurate interpretation for dog training. As the first step of this, we used inertial measurement units (IMU) to remotely detect the behavioral activity of canines. Decision tree classifiers and Hidden Markov Models were used to detect static postures (sitting, standing, lying down, standing on two legs and eating off the ground) and dynamic activities (walking, climbing stairs and walking down a ramp) based on the heuristic features of the accelerometer and gyroscope data provided by the wireless sensing system deployed on a canine vest. Data was collected from 6 Labrador Retrievers and a Kai Ken. The analysis of IMU location and orientation helped to achieve high classification accuracies for static and dynamic activity recognition.

Posture estimation for a canine machine interface based training system

Training and handling working dogs is a costly process and requires specialized skills and techniques. Less subjective and lower-cost training techniques would not only improve our partnership with these dogs but also enable us to benefit from their skills more efficiently. To facilitate this, we are developing a canine body-area-network (cBAN) to combine sensing technologies and computational modeling to provide handlers with a more accurate interpretation for dog training. As the first step of this, we used inertial measurement units (IMU) to remotely detect the behavioral activity of canines. Decision tree classifiers and Hidden Markov Models were used to detect static postures (sitting, standing, lying down, standing on two legs and eating off the ground) and dynamic activities (walking, climbing stairs and walking down a ramp) based on the heuristic features of the accelerometer and gyroscope data provided by the wireless sensing system deployed on a canine vest. Data was collected from 6 Labrador Retrievers and a Kai Ken. The analysis of IMU location and orientation helped to achieve high classification accuracies for static and dynamic activity recognition.

Balancing noise sensitivity, response latency, and posture accuracy for a computer-assisted canine posture training system

This paper describes a canine posture detection system composed of wearable sensors and instrumented devices that detect the postures sit, stand, and eat. The system consists of a customized harness outfitted with wearable Inertial Measurement Units (IMUs) and a base station for processing IMU data to classify canine postures. Research in operant conditioning, the science of behavior change, indicates that successful animal training requires consistent and accurate feedback on behavior. Properly designed computer systems excel at timeliness and accuracy, which are two characteristics most amateur trainers struggle with and professionals strive for. Therefore, in addition to the system being ergonomically designed to ensure the dogs comfort and well-being, it is engineered to provide posture detection with timing and accuracy on par with a professional trainer. We contend that providing a system with these characteristics will one day aid dogs in learning from humans by overcoming poor or ineffective timing during training. We present the initial steps in the development and validation of a computer-assisted training system designed to work outside of laboratory environments.The main contributions of this work are (a) to explore the trade-off between low-latency responses to changes in time-series IMU data representative of posture changes while maintaining accuracy and timing similar to a professional trainer, and (b) to provide a model for future ACI technologies by documenting the user-centered approach we followed to create a computer-assisted training system that met the criteria identified in (a). Accordingly, in addition to describing our system, we present the results of three experiments to characterize the performance of the system at capturing sit postures of dogs and providing timely reinforcement. These trade-offs are illustrated through the comparison of two algorithms. The first is Random Forest classification and the second is an algorithm which uses a Variance-based Threshold for classification of postures. Results indicate that with proper parameter tuning, our system can successfully capture and reinforce postures to provide computer-assisted training of dogs.

Wearable SpO2 and sleep posture monitoring system for Obstructive Sleep Apnea patients

Obstructive Sleep Apnea (OSA) affects 20% of adults in the world and is caused by a collapse of the soft tissue surrounding the upper airway, obstructing airflow. The vast majority of mild and moderate OSA patients are positional patients, which means that these patients show most of their breathing abnormalities while sleeping in supine position. For these patients positional therapy may be a simple and effective treatment solution. In this study we present a training system for positional patients to alert them when a SpO2 desaturation is occurring, or when they spend a long period of time sleeping in supine position. The alert is given to the patient initially as a smooth vibration in their wrist band, but if the condition persists, a buzzer is used for auditory indication. The system uses an accelerometer on the chest or abdomen to determine the sleeping posture and a commercial finger clip pulse oximeter to monitor heart rate and SpO2.

Stimulus Control for Semi-autonomous Computer Canine-Training

For thousands of years, humans have domesticated and trained dogs to perform tasks for them. Humans have developed areas of study, such as Applied Behavior Analysis, which aim to improve the training process. We introduce a semi-autonomous, canine-training system by combining existing research in Applied Behavior Analysis with computer systems consisting of hardware, software, audio, and visual components. These components comprise a biohybrid system capable of autonomously training a dog to perform a specific behavior on command. In this paper we further our previous computer canine-training system by the application of stimulus control over a newly-acquired, free operant behavior. This system uses light and sound as a discriminative stimulus for the behavior of a dog pushing a button with its nose. Indications of simple stimulus control of this behavior were achieved. Our pilot of this system indicates canine learning comparable to that from a professional dog trainer.

Towards a wearable system for continuous monitoring of sniffing and panting in dogs

Although numerous advances have been made in instrumental odor detection systems, these still cannot match the efficient sampling, odor discrimination, agile mobility and the olfactory acuity of odor detection dogs. A limiting step in using dogs to detect odors is the subjectivity of the translation of odor information processed by the dog to its handler. We present our preliminary efforts towards a wireless wearable system for continuous auscultation of respiratory behavior by recording internal sounds at the neck and chest by means of a commercially available electronic stethoscope to provide objective decision support for handlers. We have identified discrete features of sniffing and panting in the time domain and utilize event duration, event rate, event mean energy, and the number of consecutive events in a row to build a decision tree classifier. Since feature extraction requires segmentation of individual sniffing and panting events, we developed an adaptive method using short-time energy contour and an adaptive threshold. The performance of the system was evaluated on recordings from a Greyhound and a Labrador Retriever and achieved high classification accuracies.

Auto-adjusting mandibular repositioning device for in-home use

Obstructive Sleep Apnea (OSA) is a form of respiratory dysfunction that affects 20% of adults in the world. Among the first-line treatments that are used to mitigate the effects of OSA are continuous positive airway pressure (CPAP) and mandibular repositioning devices (MRD). Although CPAP provides a more efficacious therapy than MRDs, recent studies suggest that both are comparable in overall effectiveness due to greater patient preference and adherence to MRD therapy. In this paper, we present the Auto-Positioner, a novel add-on for MRDs that adjusts the extent to which the mandible (lower jaw) is advanced in response to respiratory signals indicating labored breathing during sleep, and to changes in sleeping position known to affect individual patients airway patency.Obstructive Sleep Apnea (OSA) is a form of respiratory dysfunction that affects 20% of adults in the world. Among the first-line treatments that are used to mitigate the effects of OSA are continuous positive airway pressure (CPAP) and mandibular repositioning devices (MRD). Although CPAP provides a more efficacious therapy than MRDs, recent studies suggest that both are comparable in overall effectiveness due to greater patient preference and adherence to MRD therapy. In this paper, we present the Auto-Positioner, a novel add-on for MRDs that adjusts the extent to which the mandible (lower jaw) is advanced in response to respiratory signals indicating labored breathing during sleep, and to changes in sleeping position known to affect individual patients airway patency.