Bottom line: storage savings are not an acceptance criterion for evidence quality. Content-aware compression makes decisions about where to spend bits; commissioning must show that those decisions preserve the operational detail required during difficult events.
\nThe Axis paper Axis Zipstream technology describes vendor algorithms that analyze video in real time and preserve selected important detail while compressing other areas more heavily. Axis reports average bandwidth and storage savings for its technology and describes features such as a dynamic region of interest.
\nAn average across scenes does not predict a specific incident. A quiet lobby can compress very differently from rain, foliage, flashing light, crowds, vehicle motion, sensor noise, or simultaneous movement across the image.
\nThe paper documents Axis technology and vendor test claims; it is not a guarantee of a fixed savings percentage or evidentiary result for any deployment. Results depend on camera model, AXIS OS version, codec, strength setting, frame rate, group-of-pictures behavior, scene, exposure, and VMS handling. Other manufacturers’ similarly named features can work differently.
\nThe following steps are DSE recommendations based on the cited source.
\nDefine repeatable incident actions at relevant distances, then record them with production lighting, shutter, resolution, frame rate, analytics, and compression. Include a quiet baseline plus the hardest credible motion and noise conditions. Retrieve the native archive and score the defined task without knowing which compression setting produced the sample when practical.
\nMeasure both quality and resource use: sustained and peak bitrate, storage per interval, recorder load, export result, and any visible artifact that affects the task. Select the lowest resource setting that consistently passes, preserve headroom for scene variation, and lock the accepted profile through change control.
\nInclude an independent reviewer who did not configure the camera. Ask that reviewer to complete the real observation or identification task from archived footage, not merely rate the image as attractive or sharp.
\nRetain the task criteria, scene script, camera and software versions, every tested configuration, native clips, blinded scores where used, bitrate traces, storage calculations, peak observations, and approval. Repeat when the scene, illumination, codec, firmware, recorder, analytics, or compression implementation changes.
\n