Key Features:
Person Characterization: Analyze videos, still images or monitor live feeds for specific identities that have been uploaded
Alert on Person(s) of Interest: Alert when a known or unknown person appears in a live stream
Vehicle Recognition: Analyze videos for specific vehicle identities
Object Detection: Search for backpacks, handbags, cars, buses, motorcycles and more, including custom object models created in the MINT module deployed with FMV
Zone Monitoring: Create user defined persistent zones for alerting in historical videos or live feeds
Identity Creation/Package: Discover and create persons of interest. Associate identities and proxies that are in the same network together and query for their presence across data
Network Discovery: Discover potential target associates through on-the-fly identity creation and querying for a correlation to a known person of interest
Collaboration: Create, share and export queries, reports, and identity packages for greater collaboration across teams
Simultaneous Real Time Monitoring: Analyze up to 10 active live streams per instance that can be private per user or shared among teams
Automatic learning: Confirm hits on persons of interest to an identity package which automatically strengthens the accuracy of the search
Model Integration: Incorporate 3rd party deep learning models such as TensorFlow or MxNet to recognize objects not already present in Mirage out of the box
Confidence Thresholding: Adjust the confidence threshold level so the system alerts on only high confidence recognition or includes lower confidence recognitions when querying with a weak or dated identity package
Color Search: Analyze historical videos for objects based on color input
Configurable Map Tile Server: Geo-correlate analysis on map of choice for secure network environments
Data Sharing & Data Privacy: Stored data in an environment can be private per user or shared among team members to facilitate collaboration
Continuous Data Import: Connect Mirage to an external data repository for continuous data ingestion
Data Export: Export thumbnail images of confirmed identities and video clips or identity package reports via docx, pdf, ppt

Key Features:
Discovery of Potential Vehicle of Interests (VOIs)
Geo-correlated Characterization to Track VOIs across missions
Comprehensive Understanding in Real-time
Integrated License Plate Recognition
Identify and Confirm Vehicles of Interest
Vehicle and Route Tracking

Key Features:
Object Detection: Automated ability to detect all classes of objects including vehicles, aircraft, ships, and other objects as needed.
Custom Object: By default, a set of pretrained objects, or vehicle types are automatically identified by Mirage. A Mirage administrator can add new custom objects to the system so that new objects can be labeled and detected over time. As analysts encounter these objects when reviewing imagery, they can either draw in bounding boxes for these objects, or reclassify existing vehicle detections to these new objects. Once there are enough labels for this new object, a new model can be trained to detect the object and Mirage will be able to identify these objects across searches. Mirage has been pioneering methods for detecting objects with only a small number of required samples as training data. Once a minimal number of samples is reached, Mirage automatically starts detecting the new object type.
Automated Classification/Categorization: Automated capabilities to classify, categorize, and identify objects using pre-defined sets of categories for any class of objects
Analyst Workflow Optimized for AI/ML human + machine teaming: Provides detection, classification and geo-correlation of objects from airborne and satellite imagery where the analyst is able to Confirm/Reclassify/Reject the image to generate training data and analyze imagery. Designed to build trust by leveraging training data and analyst confirmations to improve the accuracy of detections over time. The analyst can confirm detections as correctly classified by Mirage, or correct them. Both confirmations and corrections can be made one by one, in a batch mode from the map view, or using “quick confirm” mode. The analyst can also indicate their certainty of a confirmation assessment. These designations are included as structured data when exporting data.
Automated learning and strengthening of classification models: Delivers constantly improving mission accuracy through analyst confirmations and by monitoring model performance to identify and mitigate model drift
ELT features: Built-in tools in the UI remove haze to make images covered by clouds or haze more visible and adjust the contrast and brightness of the image. There is also a measuring tool to measure distances on the map or the length of a vehicle.
Intuitive geo-fencing and automated alerting: Users can geo-fence a zone and set persistent overwatch and alerting. The map view is optimized for seeing how a particular geofence has changed over time. Alerts can be set up to continuously monitor new imagery that comes into the platform for defined alert conditions which can then be tracked from a Mission Dashboard. Alerts can be triggered based on counts of vehicles/ships/aircraft when processing new imagery or for any object class and subset types. Multiple conditions can be set for one alert. Alerts can be pushed to SMS, email or other forms of communication.
Multiple Model Detection: Ability to analyze a single image with multiple AI/ML computer vision models to detect multiple types of objects on a single image to include aircraft, ships and vehicles
Object Specificity: Mirage has the ability to identify objects with more specificity that can attribute objects with required information, including model confidence, model version, location, date-time group, etc.
Model Performance: Ability to monitor performance of pre-trained models, visualize the amount of labeled training data and use of synthetic training data. View current model performance, compare current model to new models being generated based on analyst confirmations. Deploy and apply models across the Mirage system. View the training confusion matrix and the number of samples for each object class that went into the training data.
FI Scores: Through our Model Precision heatmaps, Mirage provides the calculation of F1 scores, including precision, and recall within each model version.
Feature Detection: Mirage has the ability to detect features, such as airstrips, uniquely shaped buildings or structures, berms, facilities, etc. over large geographic areas of a consistent type of geography using our proprietary Custom Object Detection feature called MINT.
Performance Across Geographic Areas: Mirage has the capability to train or tune models after several iterations of user-machine teaming and re-training using the models against source images in different environments and terrain.
Search for New Objects: Mirage has the ability to periodically search area for new objects or features, in addition to maintaining custody of previously identified objects or features to determine new activity with alerts.Open architecture: System that operates at speed and scale, is secure, containerized, completely open, and integratable.
Organic labeling and integration with other CV models: Enables the creation of new models for additional object classes therefore allowing analysts to characterize their environments more fully over time
Pattern of life detection: The analyst can set up alerts so that pattern of life changes or troop movements generate automatic notifications.
Geo-correlation: Mirage geo-correlates objects across data sources
Export and Import of Models: Mirage uses an open architecture whereby its models and training data can be exported to other environments and potentially other AI frameworks. Just the same, the open architecture allows Mirage to import an AI model built by another vendor or organization. The models can be in either TensorFlow, MxNet, or ONNX format for sharing neural network models. This process also includes a Quality Control step where true positive, false positive, true negative, false negative user provided confirmations generate high-quality, labeled re-training data before it is pushed to production.
Neural Network Models: Our object detectors utilize Neural Network architecture and include an automated re-training loop, driven and managed by the analysts directly in the platform to improve model performance as the platform is being used.
Export of Reports and Scenes: Scenes of interest uncovered during analysis can be captured to reports for export as GeoJSON, KML, or PPT files. These exports include all scene titles and notes with coordinates of scene boundaries, all detections with their object type, Mirage confidence percentage and confirmation status, report executive summary, and list of images contained in the report.
Open Architecture: System that operates at speed and scale, is secure, containerized, completely open, and integratable.