Data collection and retraining in edge video devices
A digital video camera architecture for updating an object identification and tracking model deployed with the camera is disclosed. The invention comprises optics, a processor, a memory, and an artificial intelligence logic which may further comprise artificial neural networks. The architecture may identify objects according to a confidence threshold of a model. The confidence threshold may be monitored over time, and the model may be updated if the confidence threshold drops below an acceptable level. The data for retraining is ideally generated substantially internal to the camera. A filter is generated to process the entire field data set stored on the camera to create a field data subset also stored on the camera. The field data subset may be run through the model to generate cases that may be used in further monitoring, training, and updating of the model.
1. A device, comprising:
an artificial intelligence logic configured to:
evaluate a validation data set with a first production detector;
use annotation of an output of the first production detector to generate a surrogate data set;
train a filter with the surrogate data set;
evaluate an entire field data set with the filter to generate a field data subset;
evaluate the field data subset with the first production detector to generate a final field data subset;
add the final field data subset to the validation data set,
updated a training data set with the validation data set;
train the second production detector with the updated training set; and
replace the first production detector with the second production detector, wherein the field data subset is run through a second production detector to generate a locally stored portion of the final field data set.
2. The device of claim 1 , wherein the annotation of the output of the first production detector is, at least in part, performed manually.
3. The device of claim 1 , wherein the annotation of the output of the first production detector is performed computationally.
4. The device of claim 1 , wherein the device further comprises an optics module and the field data set comprises video data obtained via the optics module.
5. The device of claim 1 , wherein the field data set comprises surveillance video.
6. A system, comprising:
a computational system configured to:
evaluate a validation data set with a first production detector;
use annotation of an output data of the first production detector to generate a surrogate data set; and
train a filter with the surrogate data set; and
a device, comprising:
an artificial intelligence logic configured to:
receive the trained filter;
evaluate an entire field data set with the filter to generate a field data subset;
evaluate the field data subset with the first production detector to generate a final field data subset;
add the final field data subset to the validation data set;
update a training data set with the validation data set;
train the second production detector with the updated training set; and
replace the first production detector with the second production detector, wherein the field data subset is run through a second production detector to generate a locally stored portion of the final field data set.
7. The system of claim 6 , wherein the annotation of the output of the first production detector is, at least in part, performed manually.
8. The system of claim 6 , wherein the annotation of the output of the first production detector is performed computationally.
9. The system of claim 6 , wherein the device further comprises an optics module and the entire field data set comprises video data obtained via the optics module.
10. The system of claim 6 , wherein the entire field data set comprises surveillance video.
11. A method, comprising:
evaluating a validation data set with a first production detector;
annotating output data of the first production detector to generate a surrogate data set;
training a filter with the surrogate data set;
evaluating an entire field data set with the filter to generate a field data subset, wherein a portion of the entire field data set comprises surveillance video;
evaluating the field data subset with the first production detector to generate a final field data set;
adding the final field data to the validation data set;
updating a training data set with the validation data set;
training the second production detector with the updated training set; and
replacing the first production detector with the second production detector, wherein the field data subset is run through a second production detector to generate a locally stored portion of the final field data set.