Detection Logging

The logging stage records detection events with timestamps, locations, and confidence scores for analysis and reporting.

Overview

Every matched detection is logged with comprehensive metadata, enabling historical analysis, reporting, and integration with external systems.

Log Entry Schema

Field	Type	Description
`log_id`	UUID	Unique entry identifier
`timestamp`	TIMESTAMP	Detection time (UTC)
`class_id`	INTEGER	Matched object class
`class_name`	VARCHAR	Human-readable label
`confidence`	FLOAT	Match confidence score
`camera_id`	VARCHAR	Source camera identifier
`bbox`	JSON	Bounding box coordinates
`frame_id`	INTEGER	Source frame number

Logging Pipeline

Step	Action	Output
1	Receive match result	Detection + class info
2	Enrich metadata	Add timestamp, camera, location
3	Duplicate check	Prevent rapid re-logging
4	Write to database	Persisted log entry
5	Trigger events	Webhooks, alerts

Duplicate Prevention

Avoid logging the same object repeatedly:

Parameter	Default	Description
`cooldown_seconds`	30	Minimum time between same object logs
`iou_threshold`	0.7	Spatial overlap threshold
`track_id`	-	Object tracking ID (if available)

Storage Options

Backend	Use Case	Features
PostgreSQL	Production	ACID, queries, joins
SQLite	Development	Simple, embedded
CSV	Export	Human-readable, portable
JSON Lines	Streaming	Append-only, flexible

Database Schema

CREATE TABLE detection_logs (
    log_id UUID PRIMARY KEY,
    timestamp TIMESTAMP NOT NULL,
    class_id INTEGER REFERENCES object_classes(id),
    class_name VARCHAR(100),
    confidence FLOAT,
    camera_id VARCHAR(50),
    bbox JSONB,
    frame_id INTEGER,
    created_at TIMESTAMP DEFAULT NOW()
);

CREATE INDEX idx_logs_timestamp ON detection_logs(timestamp);
CREATE INDEX idx_logs_class ON detection_logs(class_id);
CREATE INDEX idx_logs_camera ON detection_logs(camera_id);

Export Formats

Format	Extension	Use Case
CSV	.csv	Spreadsheet analysis
JSON	.json	API consumption
Parquet	.parquet	Big data pipelines
Excel	.xlsx	Business reports

Query Examples

Detections by Time Range

Filter	Description
Today	Logs from current date
Last hour	Recent detections
Custom range	Start and end timestamps

Detections by Class

Filter	Description
Single class	All logs for one object type
Multiple classes	Filter by list
Exclude classes	All except specified

Aggregations

Metric	Description
Count by class	Total detections per type
Count by hour	Hourly detection volume
Count by camera	Detections per source
Average confidence	Mean match confidence

Event Triggers

Event	Trigger	Action
New detection	Any match logged	Webhook, database write
High-value object	Specific class detected	Alert notification
Anomaly	Unusual detection pattern	Escalation
Threshold exceeded	Count > limit	Warning

Retention Policy

Strategy	Duration	Description
Keep all	Indefinite	Full history
Rolling window	30/90 days	Delete older logs
Archive	1 year	Move to cold storage
Aggregate only	After 30 days	Keep summaries

Integration

System	Method	Purpose
REST API	HTTP	External queries
Webhooks	POST	Real-time notifications
Message queue	Pub/sub	Event streaming
Dashboard	SQL	Visualization

Detection Logging

On this page