Progress: writing …

Abnormal detection

Environment

This project focus on monitoring video cameras within the lift.

What counts for “abnormal”?

There are several cases that counted as potential hazard or against general regulations:

• E-Bike: may cause hazard situations either within the lift of the building, especially fire.
• Animal: e.g., dogs without leash.

There are also cases that may imply accidents or emergencies:

• Fall down: especially elder people.

Solutions

Hardware configuration

We have one HKVision camera installed on each lift.

• Data stream: RTSP protocol.
• Transport: UDP connection.
• Multi-threading: #camera (one thread for a single camera) plus 1 (detection logics)

Computation on Server

Tested on an 8-core headless 2U server (Dell PowerEdge R540) with CentOS 7.

Computation on Edge

Tested on Raspberry Pi 4 Model B (8G) with 64-bit OS.

Visualization

Abnormal situations are sent to the server through HTTP POST with the following data specification:

• imei (str): unique serial number of the lift.
• errtye (int): error type.
• time (str): accurate to seconds.
• imgs (array): base64 encoded image list.

We also provide front end (web UI) for data visualization.

Show HTTP POST Demo.

Deployment

We provide a fully configured Docker image containing all prerequisites and the software.