Troubleshooting: Detection Performance Issues
1. Problem: A Specific Object in the Camera View Is Constantly Triggering False Detections
Solution: Exclude or Adjust the Area Where the Object Appears How to Use:
In the Active Areas (detection zones) settings, deselect or crop out the region containing the problematic object. Or build polygons around the problematic object.
Reposition or angle the camera so the offending object is no longer in the critical detection zone.
If the object is small yet triggers alarms, increase the Min Object Size to filter it out.
Why It Helps:
By excluding or modifying the area with the troublesome object, you prevent repeated false triggers.
This lets the system focus on legitimate movement, reducing nuisance alarms.
2. Problem: Time-Based Detection Schedules
Solution: Configure Detection to Be Active Only During Specific Hours How to Use:
In Rule settings, create a schedule or profile for detection hours (e.g., nights, weekends).
Disable or lower sensitivity during times of expected activity (e.g., business hours) to reduce frequent false alarms.
Save the schedule and verify that detection only runs during the specified time windows.
Why It Helps:
Prevents unnecessary alerts when people or objects are frequently moving (e.g., daytime foot traffic).
Focuses detection on high-risk periods, improving the overall signal-to-noise ratio.
3. Problem: Too Many False Alarms from Small or Irrelevant Objects
Solution: Increase the Min Object Size setting. How to Use:
Go to the Active Areas or Object Detection settings.
Set Min Object Size to a higher value (e.g., from 0 to 10 or 20).
Test if small objects (e.g., leaves, small animals) are still triggering alarms. Adjust further as needed. The size of detected objects for reference should be visible at the labels of the detected object snapshots.
Why It Helps:
By ignoring very small objects, you eliminate trivial triggers—leading to fewer false alarms and a more reliable detection system.
4. Problem: Missing Actual Objects (Underdetection)
Solution: Lower the Threshold (detection confidence). How to Use:
In the detection settings, locate the Threshold parameter.
If it’s at 0.6, try reducing it to 0.5 or 0.4.
Monitor whether the system now detects previously missed objects.
Why It Helps:
A lower threshold means the system accepts slightly less “confident” detections, ensuring fewer real objects slip through unnoticed. This might increase the false alarm rate though.
5. Problem: Excessive Alarms in Irrelevant Areas
Solution: Refine Active Areas to detect only where it matters. How to Use:
In the camera or device interface, draw or select the zones that need monitoring.
Exclude busy areas (e.g., roads, sidewalks, moving foliage).
Save changes and observe if false alarms decrease.
Why It Helps:
By limiting detection to critical zones (like doorways, gates, or restricted areas), you reduce noise from high-traffic or irrelevant regions.
6. Problem: Poor Detection in Certain Lighting or Angles
Solution: Optimize Camera Placement and Lighting How to Use:
Reposition the camera to have a clear view of the area—avoid strong backlight, reflections, or obstructions.
Use additional lighting or enable IR mode in low-light environments.
Adjust camera settings (exposure, gain) if available.
Why It Helps:
Good visibility ensures the object detection algorithm has high-quality footage to analyze, reducing misdetections caused by shadows or glare.
7. Problem: High CPU Usage or Multiple Streams Causing Missed Alerts
Solution: Monitor System Performance and Limit Overload How to Use:
If usage is too high, reduce the number of simultaneous streams or lower the resolution/bitrate.
If feasible, add additional DEKI SAFE device or split workloads across multiple devices.
Why It Helps:
Overloaded hardware can skip frames or lag, leading to missed detection's or delayed alerts. By default each DEKI SAFE device should be able to handle up to 6 full HD incoming video streams. This number can deviate based on detection model used or if different cameras from different vendors are used on one device. Balancing resource usage improves overall reliability.
By addressing each of these problems with the corresponding solutions, you’ll significantly reduce false alarms and enhance detection accuracy. If issues persist, please reach out to DEKI Support at support@deki.ai.
Last updated