Why Object Detection Works Here
80+ object classes
Detects people, cars, bicycles, animals, phones, laptops, and dozens of other everyday object classes out of the box, no custom training required.
Adjustable confidence & size thresholds
Score threshold and minimum/maximum bounding-box area are both exposed as sliders — filter out low-confidence or implausibly tiny/huge detections without touching code.
Custom labels per class
Up to four custom text labels can be mapped to specific detected classes — e.g. relabeling every detected “person” as “Dancer” for a performance piece.
Combine with motion for organic outlines
Person Tracking mode layers frame-differencing motion detection inside each detected person’s box, producing an organic blob-shaped outline instead of a rigid rectangle.
Confidence percentage on-screen
Each detection can optionally show its confidence score alongside its label — useful for tuning thresholds or just for a data-overlay aesthetic.
Export detections as CSV
Every detected object’s class, position, and confidence can be exported frame-by-frame as CSV — useful for downstream analysis of what was detected, when.
Frequently Asked Questions
What objects can this detect?
Over 80 everyday classes — people, vehicles (car, bicycle, motorcycle, bus), animals (dog, cat), and common objects (phone, laptop, bottle, chair, sports ball, and more).
Can I filter out low-confidence detections?
Yes — the Score Threshold slider filters out any detection below a chosen confidence percentage in real time.
Can I relabel a detected class with my own text?
Yes — up to four Custom Labels can be assigned to specific object classes (e.g. mapping “person” to a custom name) in the Labels settings.
Does object detection run on live video or only still images?
Both — live webcam feed, an uploaded video file, or a single still photo all run through the same detection pipeline.
