Video Task

Video detection tasks are one-time AI detections designed to simulate real-time detection scenarios. Here, you can adjust various AI and event parameters before executing them on a live stream.

What is a Video Task?

Video detection tasks are one-time AI detections designed to simulate real-time detection scenarios.

Through video tasks, you can adjust various AI and event parameters by testing detection results repeatedly, verifying your AI detection scenario plans.

Step 1: Add a New AI Video Task

1. On the Task page, click the + Task button on the top right corner.

1. Click + Advanced to show the Video Task option.

2. Click on the Video Task option.

Step 2: Configure Video Task Information

Fill out the following fields:

• Task Name: Enter a name for the task. Often this is set to be the camera's name in its video management system (VMS).

• Location:

  • Choose an existing organization and location.

  • Or click New to define a new location.

    • For new locations, the inputted coordinates will define that location's time zone.

  • Read more on location setup here.

Hint: For details on how to fill in each field, hover over the grey i tooltips.

Advanced Options (Optional)

1. CPU / GPU Decode

   Select the preferred decoding option:

   • CPU: Standard video decoding.

   • GPU: Utilizes GPU acceleration for faster processing—recommended for high-resolution or high-frame-rate video. Requires compatible NVIDIA GPU.

2. Compute Target Choose the cluster where inferencing will run:

   • Each cluster contains multiple nodes with the same configuration.

   • Large scale deployments may have clusters in different locations or networks (e.g., cluster1 with 3 nodes, cluster2 with 8 nodes.)

3. Compute Node Select the specific machine within the cluster that executes the inference task.

Once everything has been filled in, click Next on the bottom right.

Step 3: Select AI Model Detection Scenario

Scenarios are the high-level categories or that define what your camera is looking for (e.g., Vehicle Counting, PPE Detection, or Security).

Choose one or more AI detection scenarios you would like to focus on for that camera. Available options include (but are not limited to):

• Virtual Fencing

• PPE (Personal Protective Equipment) Compliance

• Fire and Smoke Detection

• Construction Activity Monitoring

• Pathway or Object Stacking Detection

• People and Traffic Counting

If the scenario you want to select is not available in the options, you can click Skip & Create to create a blank task and upload your custom scenario later on.

Otherwise, click Next to proceed.

Hint: Supported scenarios may vary depending on the software version of Observ deployed.

Step 4: Event Types and Notifications

Event Types are the specific triggers within a scenario that actually generate an alert or event log. After selecting the desired scenarios to detect, select the specific event type related to that scenario.

Hint: Here's an example. After selecting the Vehicle Counting scenario, tick Real-Time Bus Count Alert and Real-Time Truck Count Alert to isolate the detection to just buses and trucks.

Afterwards, click Create to complete task setup.

Once everything has been configured, your video task is now ready for simulation!

Navigating to Your Video Task

Unlike the AI Camera Tasks, the Video Task can be found by selecting Video. Note that this selection only appears once a Video Task has already been created.

Below is the view when clicking into your selected Video Task.

Running a Video Detection

To start running a video detection, click + New Video on the upper right of the screen to open the video details window.

Fill in the required fields and upload your video, following the specifications outlined. It is important to note that the chosen video's start time will dictate the order that events show in your records and results tab.

Once uploaded, you will see an Uploading... Status under the corresponding record in the Video List. Give the video some time to upload, which ranges from a few seconds to a few minutes depending on the video length.

When the video has finished uploading, the Status will change to Initialized. Once all the videos you've uploaded are successfully initialized, you can now start the detection by clicking on the Start Detection button.

Important: If you chose to Skip & Create the task without selecting an event in Step 3, ensure you add the event now before starting the detection.

Adding New Events

To add new events--including custom events and complex events not included in the template list in Step 3--navigate to settings on the upper right of the screen.

Under Event Type, select + Detection Event Type and select from a list of created events. In such a case, ensure that any custom events have already been created before adding them to the Video Task.

Viewing Video Detection Results

After clicking start detection, the status refreshes to reflect that it's Scheduled. Allow the task to run for the duration of the video, then refresh the site to view detected events.

Once the video is done processing, the page can be refreshed to view all the detected results. The status of the video also changes to Completed.

To review the bounding box results, select View on the corresponding video's row and replay the upload. This feature can assist the user in debugging any missed events.