Video Intelligence and Multi-Sensor Fusion
SALSA VIP is an advanced Multi-Sensor Fusion platform that integrates leading video intelligence and analytics with video management software for Public Safety, Law Enforcement, and Defense applications. SALSA VIP correlates, fuses, and aggregates data from multiple sources to detect, characterize, alert, and track vehicles of interest across hundreds of streaming traffic cameras and archived data.
The SALSA Video Knowledge Graph TM machine learning algorithm correlates sensor observations to physical objects across space and time. A confidence for the fused data is computed based on similarity of the sensed features (i.e., attributes), the provenance, and pedigree of the observations. Man-machine teaming continually improves the accuracy of the results.
Watch What Matters
Using Artificial Intelligence, Machine Learning, and Computer Vision, SALSA VIP extracts operational insights from video to generate highly accurate situational awareness. Actively search or create alert profiles for vehicles of interest based on type, make, model, color, location, time, and direction of travel, yielding high confidence results within seconds.
Vehicle Search
Search for vehicles based on a singular attribute or a combination of attributes (type, color, make, model, location, time, direction of travel).
- Search archived and live video feeds simultaneously
- Supports large number of cameras
- Highly accurate results (above 80% accuracy)
Alerts
Receive real-time alerts based on vehicle descriptions and behaviors, such as:
- Vehicle approaching at an unusual time
- Vehicle arriving/departing a location of interest
- Vehicle approaching an unauthorized area
- Unusual traffic volume in an area for the time or day
- Vehicle parked at an unusual location
Pattern of Movement Analysis
SALSA VIP provides :
- Vehicle’s last known locations, top destinations, and patterns of travel
- Vehicle arrival and/or departure times
- Total number of vehicles by day-of-the-week, hour-of-the-day, month-of-the-year
Semandex Video Knowledge Graph™
The VKG enriches observations derived from multiple sources to amplify the context of video sensor observations and enable high performance entity resolution and tracking.
- Links previous exploitation and source video with other source reporting
- Enables a higher level of analysis for future activity anticipation
SALSA VIP Specifications
Video Input
- MPEG-2, Simple, Main, and 422P Profile up to High Level
- MPEG-4 AVC/H.264, Baseline, Main, and High Profile Up to Level 4.2 HD
Video AI engine
- Classification
- Base Mode: bounding box (up to 15 fps)
- High Fidelity Mode: Contour (up to 5 fps)
- Object re-identification (REID)
- Signature computation from live and archived video
- High confidence matches with up to 30-degree perspective difference across cameras
- 80% accuracy with fair lighting conditions using pre-trained model
- 90% accuracy with offline incremental training and man-machine feedback
SALSA VIP is designed to ingest and manage data from varied sensors including:
- RFID sensor observations
- RF sensor observations
- Automatic License Plate Reader (ALPR) detections
- Text-based reports
SALSA VIP fuses data from multiple sources and generates relationship analysis to uncover associations between the objects in video observations, people, communications activities, places, events and transactions.
- Fusion of live streaming and archived observations
- 100’s of live video feeds and 1000’s of hours of archived video
- SALSA Video Knowledge Graph ™ links observations to physical objects across space and time
- Multi-source observations are synthesized taking into account the confidence values of each sensor detection
- Man-machine teaming input enables continually and automatically improve the accuracy of the algorithm
- Search and alert on live streaming and archived observations
- Software architecture and container (docker) packaging enables flexible and scalable deployment on cloud and on-premise environments
Learn More
Rapid Deployment
SALSA VIP can be deployed on-premises hardware and is capable of integrating into a cloud architecture or operating as a stand-alone “Roll on/Roll off” capability. SALSA VIP containerized software package is easy to set up and manage. SALSA VIP supports immediate integration with IP-based network cameras and sensors, producing data in standardized and interoperable formats. Software APIs allow for configuration to ingest a variety of data feeds and sensor types.
Effective Scalability
Multiple software containers can be installed on any number of servers, from portable transit cases, for local coverage and data centers for regional coverage, to enterprise deployments federated across different regions, for global coverage. From back-packable high-performance computing resources used in tactical operations to on-premise high-performance computing clusters.
Edge Processing Ready
The Edge Processing Units (EPUs) provide fully operational video analytics in environments where there is limited power and bandwidth to transmit HD video. Each EPU is capable of computing vehicle signatures at a bit rate of approximately 25 Kbps per detected vehicle and offers ultra-low power high-density inference processing of approximately 2 Watts.
SALSA VIP EPUs turn a conventional camera into a “smart camera” that can be exploited as close to the sensing location as possible, while transmitting back only the minimal required information for a back-end processor to index, analyze, and cross-correlate, providing the information needed for analyst to act. The EPU provides the vehicle classification, color detection and vehicle/track characterization in real-time to a remote distributed search sub-system.
EPU Technical Specifications
- NVIDIA® Jetson™ Xavier NX embedded AI-ASIC Inference System
- Low object signature bit rate (approx. 25 Kbps) per detected object
- Low power consumption (approx. 40 Watts)
- PoE adaptable
- Fanless compact design (5.7” W x 4.6” L x 2.0″ H)
- Lightweight (3.5 lbs)
- Rugged Case
