Trap Counting in FishVue Interpret

This guide explains how to use FishVue Interpret version 3.0.9 to accurately count objects using the AI annotation interface.

It assumes your footage has already been processed with the Trap Counting model in FishVue AI.

Step 1: Load the AI Annotation in Interpret

1. Open the Video window with your AI annotation loaded.

2. Make sure annotations are enabled:

   • Click the magnifying glass icon to the left of the video timeline for the window where you want to count detections.

3. At the bottom-right of the window, click the right arrow to expand the AI Options panel.

Step 2: Set the “Merge Events” Value

1. In the AI Options panel, find the Merge Events slider.

2. Drag the slider as high as possible while still staying below the expected rate at which new objects appear.

Goal:

• Avoid merging different objects into a single detection.

• Avoid splitting one object into multiple detections.

Step 3: Filter by the Correct Class

1. In the AI Options panel, uncheck any classifications that are not the object type you want to count.

This ensures only detections for your target object class are shown and counted.

Step 4: Define a Detection Filter Area

1. Right-click on the video window and select Create Detection Filter Area.

2. Draw and adjust the area so that:

   • It is as small as practical.

   • It is centered where the object always passes through once.

   • The object is clearly visible within this area.

This area is your counting zone.

Step 5: Configure the Detection Report

1. At the top of the Video Control Window, with the AI Options expanded, click Create Detection Report.

2. In the Detection Report window:

   • Optionally set Start Time and End Time if you only want to count objects in a specific time range.

   • Leave these fields blank to count objects across the entire dataset.

Step 6: Generate Object Counts

1. In the Detection Report window, click Create Detection Snapshot.

2. The tool will calculate how many individual objects passed through your detection filter area within the selected time window.

These counts are based on objects entering the defined zone, using your chosen class filters, merge settings, and time range.

Notes:

• In the Detection Report window, you can click any event row in the snapshot to jump the video directly to that detection.

• Important limitation: The system can’t reliably tell multiple objects apart if they appear in the detection zone at the same time (it only separates them based on gaps between detection clusters). For accurate counts, make sure there is only one countable object in the detection zone at any given moment.