Identifying anomalous activity from thermal images

1. A method comprising:

training, by a computing system, an autoencoder neural network to generate a first set of codes representative of a first set of thermal video images of activities of a user in an environment, wherein:

the activities of the user represent routine behaviors of the user in the environment,

the autoencoder neural network comprises an encoder and a decoder, and

training the autoencoder neural network comprises:

applying the encoder to encode the first set of thermal video images of the user to generate the first set of codes;

applying the decoder to decode the first set of codes to reconstruct the first set of thermal video images of the activities of the user; and

determining reconstruction errors of the reconstructed first set of thermal video images of the images of the activities of the user;

categorizing, with an unsupervised machine-learning algorithm, the first set of codes into a set of clusters;

after categorizing the first set of codes into the set of clusters, determining, with the autoencoder neural network, a code representative of a second set of thermal video images of an activity in the environment;

determining, by the computing system, that the code is an anomalous code based on the code not being associated with any cluster in the set of clusters; and

based on the anomalous code, performing, by the computing system, an alert action.

2. The method of claim 1 , wherein performing the alert action comprises:

prompting, via a device associated with the user, a response to indicate whether the activity is an actionable event.

3. The method of claim 2 , wherein the response is supplied, via the device, by the user or by a trusted associate of the user.

4. The method of claim 1 , wherein performing the alert action comprises performing, by the computing system, the alert action based on a cluster from the set of clusters, wherein the cluster corresponds to routine behavior.

5. The method of claim 4 , wherein performing the alert action based on the cluster comprises:

prompting, via a device associated with the user, a first response from the user or trusted associate of the user to identify a first name of a first activity; and

labelling, with the first name of the first activity, a first cluster that corresponds to the first activity.

6. The method of claim 5 , further comprising:

determining, in the first cluster, codes labelled with a second name; and

separating codes labelled with the first name from codes labelled with the second name by splitting the first cluster into two clusters.

7. The method of claim 5 , further comprising:

determining a second cluster that corresponds to a second activity with a second name;

determining that the first name and the second name are similar; and

combining the first cluster and the second cluster into one cluster corresponding to the first activity with the first name.

8. The method of claim 5 , wherein performing the alert action based on the cluster only occurs for a cluster containing a minimum number of codes.

9. The method of claim 1 , wherein performing the alert action comprises prompting, via a device associated with the user, a response to indicate whether the activity is an actionable event, wherein the actionable event is a fall of the user.

10. The method of claim 1 , wherein the first set and the second set of thermal video images are received from one or more thermal imaging cameras in the environment.

11. The method of claim 1 , wherein the set of clusters is a plurality of clusters corresponding to a plurality of routine behaviors of the user in the environment.

12. The method of claim 1 , further comprising:

determining that the code is not the anomalous code; and

classifying the code into a cluster from the set of clusters, wherein the cluster contains one or more codes that share one or more characteristics with the code.

13. A system comprising:

one or more memory units; and

one or more processors in communication with the memory units, wherein the one or more processors are configured to:

train an autoencoder neural network to generate a first set of codes representative of a first set of thermal video images of activities of a user in an environment, wherein:

the activities of the user represent routine behaviors of the user in the environment,

the autoencoder neural network comprises an encoder and a decoder, and

the one or more processors are configured to, as part of training the autoencoder neural network:

apply the encoder to encode the first set of thermal video images of the user to generate the first set of codes;

apply the decoder to decode the first set of codes to reconstruct the first set of thermal video images of the activities of the user; and

determine reconstruction errors of the reconstructed first set of thermal video images of the images of the activities of the user;

categorize, with an unsupervised machine-learning algorithm, the first set of codes into a set of clusters;

determine, with the autoencoder neural network, a code representative of a second set of thermal video images of an activity in the environment;

determine that the code is an anomalous code based on the code not being associated with any cluster in the set of clusters; and

perform, based on the anomalous code, an alert action.

14. The system of claim 13 , wherein the one or more processors are configured to:

prompt, via a device associated with the user, a response to indicate whether the activity is an actionable event.

15. The system of claim 13 , wherein the one or more processors are configured to:

prompt, via a device associated with the user, a first response from the user or trusted associate of the user to identify a first name of a first activity corresponding to a first cluster from the set of clusters; and

label, with the first name of the first activity, the first cluster.

16. The system of claim 15 , wherein the one or more processors are configured to:

determine, in the first cluster, codes labelled with a second name; and

separate codes labelled with the first name from codes labelled with the second name by splitting the first cluster into two clusters.

17. The system of claim 15 , wherein the one or more processors are configured to:

determine a second cluster that corresponds to a second activity with a second name;

determine that the first name and the second name are similar; and

combine the first cluster and the second cluster into one cluster corresponding to the first activity with the first name.

18. The system of claim 13 , wherein the one or more processors are configured to:

determine that the code is not the anomalous code; and

classify the code into a cluster from the set of clusters, wherein the cluster contains one or more codes that share one or more characteristics with the code.

19. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause at least one processor of a computing device to:

train an autoencoder neural network to generate a first set of codes representative of a first set of thermal video images of activities of a user in an environment, wherein:

the activities of the user represent routine behaviors of the user in the environment;

the autoencoder neural network comprises an encoder and a decoder, and

the instructions that cause the at least one processor to train the autoencoder neural network comprise instructions that, when executed, cause the at least one processor to:

apply the encoder to encode the first set of thermal video images of the user to generate the first set of codes;

apply the decoder to decode the first set of codes to reconstruct the first set of thermal video images of the activities of the user; and

determine reconstruction errors of the reconstructed first set of thermal video images of the images of the activities of the user

categorize, with an unsupervised machine-learning algorithm, the first set of codes into a set of clusters;

determine, with the autoencoder neural network, a code representative of a second set of thermal video images of an activity in the environment;

determine that the code is an anomalous code based on the code not being associated with any cluster in the set of clusters; and

perform, based on the anomalous code, an alert action.