Resources
Glossary
60 terms to help you understand your video analytics project.
YOUR PROJECT VOCABULARY
60
terms to turn images into understanding.
IA · VMS · RTSP
Streams · Data · Use cases
A
AI Act
The EU regulation on artificial intelligence. Obligations depend on factors including the organisation’s role and how the system is used.
AI model
A system trained on data to perform a task. Its performance also depends on the conditions in which it is used.
Alert
A notification issued when a rule is met. An operator reviews it and decides on the response specified by procedures.
Algorithmic video analysis
Automated analysis of video to identify objects, movements or situations. Technical availability does not determine whether a use is legally permitted.
Anonymisation
A process intended to prevent a person from being identified. Blurring a face alone does not guarantee that all the data is anonymous.
API
An interface that lets software exchange information. Its functions, formats and authentication mechanisms need documentation.
Audit log
A history of actions recorded by a system. It helps review access, changes and operations according to which events are actually logged.
B
Bandwidth
The transmission capacity available on a network. It should be compared with the combined data rate of streams and other uses.
Biometric identification
Identifying a person using biological or behavioural characteristics. It is different from counting human figures.
Blurring
A visual modification that obscures image details. When it is applied determines which images remain accessible earlier in the pipeline.
C
CNIL
The French data protection authority. Its publications help explain the rules applicable to data processing.
Codec
A method for encoding and decoding images. The selected codec must be supported throughout the processing chain.
Confidence threshold
The minimum model score required to retain a detection. Raising it may reduce some alerts but also exclude useful events.
Counting
Measuring the number of objects or people according to a defined rule. Counting crossings is not necessarily the same as counting unique visitors.
Crowd density
The estimated number of people per unit of area. It differs from the total number of visitors over time.
D
Detection
Finding an object or situation in an image. A detection becomes an alert only when the configured rules require it.
Digital twin
A digital representation of a real place or system. Data can feed it to track its state or operation.
DPIA
A data protection impact assessment describes the processing and evaluates risks to people. It documents the measures intended to address those risks.
DPO
Data protection officer. This role advises and supports an organisation on its data protection obligations.
E
Edge
Processing close to the source of the data. It can run on a local server or an on-board device.
Exit and portability
The ability to recover data and switch solutions under defined conditions. Formats, timeframes and responsibilities should be specified.
F
False negative
A target event missed by the system. Evaluating it requires labelled events and a defined test method.
False positive
An alert triggered when the target event did not occur. Its rate should be measured on representative scenes using a defined counting method.
Footfall
The number of passages observed over a period. Entries, exits and possible duplicate counts should be specified.
Frames per second
The number of frames transmitted or analysed each second. The useful rate depends on object speed and the scenario.
G
GDPR
The General Data Protection Regulation. It governs personal data processing within its scope.
GPU
A processor suited to many parallel calculations. It accelerates certain image-analysis model operations.
H
Heat map
A visual representation of how a measurement is distributed across a space. It can highlight the busiest areas over a defined period.
High availability
An architecture designed to limit service interruptions. It needs to be defined and tested for the components actually deployed.
I
Inference
Using a trained model to analyse new data. It is separate from the model’s training phase.
Integration
A connection between systems to exchange data or commands. Available functions depend on the connector and software versions.
K
Kafka
A platform for publishing and processing streams of events. An integration must define the exchanged messages and access permissions.
L
Latency
The time between an event and its presentation in the system. Networking, processing and display can each contribute.
M
Metadata
Information describing an image or event. A position, category and timestamp are examples.
MQTT
A messaging protocol based on publishing and subscribing to topics. It can carry events to another application.
O
Object tracking
Associating successive detections within a sequence. It does not necessarily mean that a person’s identity is known.
Occupancy
An estimate of the people or objects present at a given time. It differs from a day’s total crossings.
On-board computing
Computing performed inside a vehicle or mobile device. Power, temperature, vibration and connectivity need assessment.
On premises
Deployment within the customer’s environment. Location does not replace access controls and other security measures.
ONVIF
A set of interoperability specifications for security devices. Conformance is checked by model and profile, not by brand alone.
P
Pilot
A limited deployment to assess a requirement under real conditions. Success criteria should be defined before starting.
PoE
Electrical power delivered through a compatible Ethernet cable. Available power must match device requirements.
Precision
The proportion of retained detections that actually match the target. This metric must be distinguished from recall.
PTZ
A camera that can pan, tilt and zoom. Changes in framing must be considered when configuring analytics.
R
Real time
Processing data as it arrives. It does not mean that there is no delay at all.
Recall
The proportion of target events that the system actually finds. Results should also be assessed alongside false positives.
Region of interest
The part of an image to which analytics are applied. Its boundary can target a specific passage, platform or perimeter.
Resolution
The number of pixels making up an image. Object size in the image and framing also matter for analytics.
Retention period
How long a category of data is kept. It is defined by the purpose and applicable rules, not just storage capacity.
Retrospective search
Searching footage that has already been recorded. It helps find relevant sequences using criteria and a time range.
RTSP
A protocol used to control media streaming sessions. A valid RTSP address and permissions can provide access to a camera stream.
S
Scenario
A combination of detection, area and triggering rules. It turns an operational requirement into a configuration.
Security supervision platform
Software that brings together events from multiple security systems. Here, it does not mean a hypervisor used to host virtual machines.
Server
Equipment providing computing or storage services. Sizing depends on the streams and processing actually required.
T
Thermal camera
A camera sensitive to thermal infrared radiation. Its image shows thermal contrasts rather than a conventional visible-light scene.
U
Urban control room
An urban surveillance and coordination centre. Operators review footage and coordinate responses under established procedures.
V
Video stream
A sequence of images transmitted by a camera or video application. Resolution, compression and frame rate affect its quality and bandwidth.
Virtual line
A line drawn in the image to count or detect crossings. Its position and direction are configured for the scene.
VLAN
Logical segmentation of a network. It helps organise traffic alongside appropriate filtering rules.
VMS
Video management software. It centralises tasks such as viewing, recording and searching camera footage.