Background
Bandwidth issues pose a significant obstacle to quickly accessing relevant video content. Images may contain valuable information, but extracting germane details requires proper application of processing techniques to the correct data. Exhaustive processing is expensive in terms of time and processing - due to the sheer size of video content - and is unlikely to succeed because many algorithms can produce false results when applied to unsuitable data. For example, running facial recognition processing on an image with no visible faces is pointless at best and may produce misleading results.
Analysis
Timed video metadata provide a mechanism with which to categorise and index video content. For example, location data can quickly find images near a point of interest, and orientation data identify which video frames can see an object of interest at a nearby location. Other examples include sensor data that can detect movement to identify when animals are visible to a wildlife camera, and accelerometer data that indicate when and how a vehicle has collided with an obstruction.
Metadata typically comprise around 1% of their associated media data (video and audio) content. Recording such metadata in a separate sidecar file enables faster access by eliminating overheads required to distinguish these data from the other 99% of the video file.
Constructing a metadata index from a sidecar file requires no access to its associated video file, so privacy and security policies can be implemented simply by using filing system permissions. Access to an associated video file is only required in order to retrieve relevant video content which has been identified by searching the metadata index.
In addition, data transfer bottlenecks can be mitigated by limiting downloads to only include relevant video files identified using metadata search results. For example, video file downloads from a wildlife camera can be restricted to those where movement was detected near a nest or burrow in the video frame. This approach is particularly valuable for low-cost devices where communication speed and on-board storage capacity may be limited.
Benchmarks
Key metrics identified for video metadata search are:
- Access time required to read all timed metadata content for indexing.
- Access time required to read a single video frame or a short video clip for retrieval of results.
Benchmarks can be calculated by comparing times required to selectively access video content based on metadata search results with those required to access the same video content without guidance from search results.
Background
Bandwidth issues pose a significant obstacle to quickly accessing relevant video content. Images may contain valuable information, but extracting germane details requires proper application of processing techniques to the correct data. Exhaustive processing is expensive in terms of time and processing - due to the sheer size of video content - and is unlikely to succeed because many algorithms can produce false results when applied to unsuitable data. For example, running facial recognition processing on an image with no visible faces is pointless at best and may produce misleading results.
Analysis
Timed video metadata provide a mechanism with which to categorise and index video content. For example, location data can quickly find images near a point of interest, and orientation data identify which video frames can see an object of interest at a nearby location. Other examples include sensor data that can detect movement to identify when animals are visible to a wildlife camera, and accelerometer data that indicate when and how a vehicle has collided with an obstruction.
Metadata typically comprise around 1% of their associated media data (video and audio) content. Recording such metadata in a separate sidecar file enables faster access by eliminating overheads required to distinguish these data from the other 99% of the video file.
Constructing a metadata index from a sidecar file requires no access to its associated video file, so privacy and security policies can be implemented simply by using filing system permissions. Access to an associated video file is only required in order to retrieve relevant video content which has been identified by searching the metadata index.
In addition, data transfer bottlenecks can be mitigated by limiting downloads to only include relevant video files identified using metadata search results. For example, video file downloads from a wildlife camera can be restricted to those where movement was detected near a nest or burrow in the video frame. This approach is particularly valuable for low-cost devices where communication speed and on-board storage capacity may be limited.
Benchmarks
Key metrics identified for video metadata search are:
Benchmarks can be calculated by comparing times required to selectively access video content based on metadata search results with those required to access the same video content without guidance from search results.