The Security Event

25-27 April 2023
NEC, Birmingham

News

Security Cameras

28 Mar 2022

Reducing false alarms with an improved object detection algorithm

Reconeyez UK Stand: 4/D50
Object detection

Object detection in the security industry

Reconeyez system uses object detection technology to detect the presence of persons and vehicles on alarm photos and to filter out false alarms. Reconeyez deep learning algorithms filter out 95% of false alarms in urban environments on average. By supporting additional features such as Define alarm area on photo and arming Schedules for devices, Reconeyez provides monitoring stations the tools to improve alarm monitoring efficiency and performance.

The main goal  of object detection is to identify and visually mark the objects on alarm photos (Auto alarms and Auto dismissed). In the security industry object detection helps to detect unwanted human intruders or cars, monitor environmental changes and track fly-tipping.

Object detection results

There are 4 possible results for the object detection:

  • True Auto alarm: or True Positives are the cases where an object (e.g person) has been predicted as positive and it indeed is a person on the alarm photo.

  • False Auto alarm: or False Positives are the cases that have been predicted as positive but they do not have persons on the alarm photo.

  • True Auto dismissed: or True Negatives are the cases that have been predicted as negative and they indeed do not have a person on the alarm photo.

  • False Auto dismissed: or False Negatives are the cases that have been predicted as negative but they have persons in the alarm photo.

All of the 4 outcomes listed above have different business values. Let’s continue with the analogy of the algorithm that is trying to identify persons in the alarm photo. Under ideal circumstances the object detection algorithm will detect all alarms with persons on the photos as Auto alarms and all alarms with no persons on the photos as Auto dismissed.

Configurable Autoalarm threshold

Object detection probability threshold allows users to manage the number  of Auto alarms. As no Object Detection algorithm  is 100% accurate, the user needs to configure an Auto alarm threshold for each object that would suit their needs.  Lower Auto alarm threshold would result in  more Auto alarms with potentially more false positive alarms. Higher Auto alarm threshold would result in less Auto alarms, but can also result in some alarms with lowered probability object detection (eg. “person” 45%) being discarded as Auto dismissed. It is generally advised not to lower the Auto alarm threshold for any object lower than medium. Also this configurable option allows to ignore some of the detected objects completely by disabling appropriate objects (eg. “bicycle”) from being detected by object detection algorithm and consequently from being classified as an Auto alarm.

Define alarm area on photo

Define alarm area on photo feature allows to determine an area of interest on the given detector’s alarm photo. This defined area will cause the object detection algorithm to only detect objects  from inside this area and ignore all objects from outside of this area. This handy tool will allow to ignore parts of the alarm photo, where there is constant movement, eg. a busy street or a parking lot in the background.

View all News
Loading

The Safety & Security Event Series

  

The Security Event Logo   The Fire Safety Event   The Health & Safety Event The Work Place Event   National Cyber Security Show