Ensuring security for remote sites often requires reliance on 4G mobile networks as the only available means of connection. In such situations, it becomes crucial to minimize the bandwidth usage of cameras in order to stay within the data limits set by the internet service provider.
With 3dEYE, smart object detection can be enabled on any ONVIF-compatible camera, regardless of the video stream resolution, allowing for automatic monitoring of remote sites using smart alerts. Under good lighting conditions, object detection can work effectively even with objects as small as 10 pixels wide.
To achieve the best results at low resolutions, the following precautions should be taken:
- Proper camera positioning to optimize viewing angle, which will improve AI accuracy and allow for lower resolutions to be used.
- Ensure optimal camera positioning without any objects obstructing the lens, such as trees.
- Avoid top-down views as they may result in poor performance for object recognition.
- The field of view of the camera is closely related to the size of the lens, with larger lenses providing smaller fields of view. To reliably detect objects at greater distances, narrow field-of-view cameras are recommended.
- Although cameras set up at greater heights (>8 feet) may provide larger fields of view, they may not effectively capture people's faces or clothing, resulting in less effective AI analytics.
- Installing the camera on a stable foundation to prevent swaying and as a result false motion detection.
- Proper illumination to improve contrast.
- It is recommended to have well-lit areas in the monitored area, avoiding setting up the camera in dark-shaded areas pointing towards brightly lit areas.
- It is advisable to utilize the WDR (wide-dynamic range) function on the camera for scenes that encompass both well-lit and dark areas. However, caution should be exercised when using WDR, as some cameras may introduce motion blur in moving objects when high values of WDR are employed. Therefore, it is recommended to use lower values for WDR to mitigate this issue.
- Overall, overexposed or underexposed camera images (capturing too much or too little light) can result in decreased performance of analytics.
- The camera shutter speed should be high to minimize motion blur.
- Shutter speed is the length of time the camera's shutter is open, controlling the amount of light reaching the sensor. Slower shutter speeds can result in the blur of moving objects. 1/100 or faster shutter speeds are recommended.
- Shutter speed is the length of time the camera's shutter is open, controlling the amount of light reaching the sensor. Slower shutter speeds can result in the blur of moving objects. 1/100 or faster shutter speeds are recommended.
- Lower FPS (frames per second) is acceptable if fast-moving object detection is not necessary, as reducing FPS from 12 to 3-5 frames per second can save 30-40% of bandwidth or even more under certain conditions.
- Lower FPS settings can still produce reliable monitoring results without consuming excessive network and storage resources, if fast-moving object detection/evaluation is not required (e.g. in a construction site).
- Many camera devices perform more reliably at lower FPS settings, due to reduced CPU consumption.
- Setting FPS to extremely low values, such as 1 or less, is not recommended, as it does not provide significant bandwidth savings compared to 3 FPS, but it may result in poor user experience and lower object detection accuracy.
Here is an example of a camera configuration that can still produce sufficient image quality for reliable object detection at a bandwidth usage of under 200 Kbps:
From shorter distances even people inside a moving vehicle are detected by this camera:
At larger distances the AI reliably distinguishes between moving and stationary vehicles under high traffic conditions:
If the scene exhibits a high level of contrast and the objects are clearly distinguishable from the background, it may be possible to further reduce the resolution, resulting in a lower bitrate, while still achieving satisfactory object detection results. For instance, a camera set to a resolution of 640x360 pixels at a frame rate of 5 frames per second can serve as an example.