People-counting systems can help a retailer understand when visitors arrive, how traffic changes through the day, and whether staffing matches demand. When counts are combined with suitable transaction data, they can also support conversion analysis. The value comes from consistently measured information and careful interpretation, not simply from installing a camera with an analytics checkbox.
This article revisits an archived comparison of six counting solutions. It describes the original reviewers' office tests and the equipment versions available to them at that time. These were not new tests performed by Smart Link Academy, and the results are not a current product ranking or a guarantee for a different building.
What business questions can counts answer?
A useful counting system distinguishes entries from exits and records when crossings occurred. That supports hourly traffic profiles, comparisons between days, and identification of busy periods. Managers can use the information to investigate staffing, queueing, merchandising, or changes in visitor flow.
Counting alone does not establish why sales changed. A conversion calculation needs a defined denominator and a compatible transaction measure. For example, transactions divided by counted entries is not necessarily the percentage of individual shoppers who bought something: groups, repeat entries, employees, and multiple purchases can affect the interpretation.
Automating collection can reduce the effort of manually observing many stores, but it does not eliminate human judgment. Staff still need to define the metric, configure the system, check its accuracy, and investigate anomalies. A consistently biased count can produce a convincing-looking but misleading report.
Three architectures in the original comparison
The first category was the dedicated counter. Its main job was to turn a scene into numerical traffic data. Some devices stored counts locally and sent them to a server on a schedule. They did not necessarily provide an independent video archive, even if their housing resembled a surveillance camera.
The second category was a surveillance camera with embedded counting analytics. This could combine ordinary video recording with visitor statistics. In the tested versions, some functions were included while others required an additional license. Feature availability must be checked for the exact model, firmware, and license configuration.
The third category was server-based analytics. Software processed a compatible camera stream, stored results, and presented reports. Integration with point-of-sale information could add business context. This approach can require more computing and administration, but it is not universally the most expensive, and it cannot be assumed to work with every camera.
| Architecture | Main strength | Planning question |
|---|---|---|
| Dedicated counter | Focused traffic measurement | Does it retain counts during outages, and is video available for validation? |
| Camera-based analytics | Video and statistics from one device | Which functions and licenses are included? |
| Server-based analytics | Central processing and reporting integration | Are stream compatibility, server capacity, and ongoing support adequate? |
How the original tests were arranged
A relatively straightforward corridor
The reviewers mounted the cameras and counters in a row on an office-corridor ceiling, 2.90 meters above the floor, approximately 9.5 feet. The lenses pointed vertically downward. A virtual counting line was placed near the middle of each image, and people passed through the observed area.
Manual observers counted entries and exits separately. The reviewers compared those reference totals with the device reports. Most products performed relatively well in this arrangement. The Hikvision sample was the notable exception: the reviewers observed one person being counted as two or three.
They suggested that mounting height and the absence of an object-size adjustment in the tested counting interface might explain that behavior. The test did not establish a single proven cause, and the camera could not be mounted higher in that corridor.
An elevator lobby with waiting and crossing traffic
The second arrangement was intentionally more difficult. Cameras looked down over elevator entrances from 3.40 meters, approximately 11.2 feet. People emerging from an elevator had little distance in which to become a stable tracked object before reaching the counting line.
Moving the line farther away introduced a different problem: people traveling along the reception corridor could enter the counting area. People leaving the office also waited for an elevator within view, sometimes remaining near the line for an extended period. Waiting, turning, crossing, and overlapping movement made this scene substantially different from a simple doorway crossing.
The original article reported that industry participants considered the location unfavorable for counting. The reviewers tested it anyway because it represented a real installation challenge. That decision makes the comparison useful as a lesson in scene design, but it does not make the two arrangements controlled equivalents.
Megacount VideoCount 3D ASSIS
The Megacount device was a specialized visitor counter. In the reviewed configuration it converted the observed video into analytics rather than maintaining a conventional video archive. It used IP networking and PoE power, and the supplied package was described as including a PoE injector.
Counts accumulated on the device and could be transferred to an FTP server at configured times. Supplied software provided a way to view the statistics. Local accumulation meant a constant computer connection was not required for every count, although actual outage retention and later recovery still depend on storage capacity and implementation.
The configuration offered independent entrance and exit lines, object tracking overlays, and image adjustments such as brightness, exposure, and contrast. The original review described both downward and angled installation options. Those options still require compliance with the device's calibration and placement requirements.
It also produced movement tracks and activity heat maps. Here, “heat map” means a visualization of where activity occurred, not a thermal measurement. Such views can help examine paths and frequently visited areas, provided the underlying detections are credible.
For the corridor, the narrative reported 98.1% for entry and 95.65% for exit under its stated accuracy terminology. The elevator-lobby test did not produce a satisfactory count. The reviewers configured the unit themselves and did not use the remote calibration service described as available at the time, an important limitation when interpreting the outcome.
CountBOX STD-1
The CountBOX STD-1 was also a dedicated counter. It did not independently maintain the same kind of video archive as a general-purpose recorder, but the source described an RTSP video stream that compatible external equipment could record. This distinction matters when planning manual validation of counting events.
Setup involved drawing a virtual line, assigning the direction for entry and exit, and setting the expected object size. The reviewer described this as a relatively simple interface. Simplicity can reduce setup effort, but it does not remove the need to verify placement, lighting, and representative traffic.
The system stored entry and exit data in a database and offered reports by hour, day, week, month, and year. Web access, PoE power, and scheduled FTP export were described. The source also noted that individual passages, as well as totals, were recorded in the data.
The corridor narrative reported 98.8% for entry and 97.8% for exit. The elevator-lobby arrangement was unsuccessful. The contrast between these scenes shows why a favorable doorway demonstration should not be treated as proof that the same configuration will work where people stop, turn, or queue.
Hikvision DS-2CD4A26FWD-IZHS
The Hikvision sample was a Full HD surveillance camera with a motorized 2.8–12 mm varifocal lens and autofocus. The archived description listed several analytics functions, including face detection, zone entry and exit events, and plate-related capability. Those historical feature claims are not confirmation of every regional version or firmware configuration.
For visitor counting, the tested interface offered a virtual line and a direction setting. The reviewers did not find a way to set object size in that interface. In the lower corridor installation, one person could be split into multiple detections, producing severe overcounting.
At the higher elevator-lobby mounting position, the narrative reported an accuracy figure of 88%, without providing separate entry and exit values in that passage. The authors considered this noteworthy relative to some other products' larger discrepancies in the difficult scene.
That result does not establish that difficult scenes improve this model in general. Height, framing, object scale, traffic pattern, and calibration all changed between arrangements. The useful lesson is to test the intended installation geometry rather than assume a product's relative ranking remains fixed.
Smartec STC-IPMX3193A
The Smartec camera used a ceiling-oriented fisheye view approaching 180°. Its broad downward image helped show the elevator and reception arrangement. A wide view can be useful for understanding a scene, but it does not guarantee that no obstruction or blind area exists, nor that every person occupies enough pixels for accurate counting.
The source described a basic VCA Presence package for detecting objects in a monitored zone and optional licensed analytics for additional functions. These included size-based classification, direction, stopping or dwell behavior, and object counting. The counting function therefore needed to be considered together with its license.
Configuration was more involved than the simpler line-and-direction interfaces. Object-size calibration was available, and the camera could display tracked objects and their paths. The reviewers emphasized the need to understand the settings and follow setup guidance.
For the corridor, the narrative reported 96.4% for entry and 95.2% for exit. The elevator-lobby test did not provide satisfactory counting. The source connected measurement quality with calibration, line placement, and the area used for counting. A broad fisheye overview did not by itself solve the tracking problem.
Acumen AiP-P25V-45Y2B
The Acumen unit was an outdoor dome camera with infrared illumination. The archived specifications described recording up to 1080p at 30 fps, autofocus, a 2.8–10 mm varifocal lens, and a quoted minimum illumination of 0.015 lux at F1.4. Those figures describe the reviewed version and are not a new independent specification test.
Four simultaneous streams were listed. Multiple streams can provide different quality levels for different clients, but transmitting all of them does not automatically save bandwidth. The source also listed face detection, traffic-light-color recognition, and visitor counting among the analytics features.
The counting interface used a line and direction setting, with crossings assigned to entry or exit. The narrative reported 92.6% entry and 91.3% exit for the corridor. In the elevator lobby, it reported 83.1% entry and 60% exit.
The large difference between directions in the difficult scene is operationally important. An apparently reasonable entry count does not establish a reliable exit count or occupancy estimate. Validate the two directions separately when the business metric depends on both.
Synesis Cassiopeia: server-based software
The original comparison used a demonstration version of the Cassiopeia analytics software. It processed camera video according to configured rules, collected statistics centrally, and presented the results in graphical reports. The approach shifted the analysis workload from a dedicated counting device to a server.
The software was described as detecting and tracking human heads in a downward-facing view. Its intended purpose was to distinguish people from shadows, reflections, merchandise, and interior objects. Those were design goals and vendor claims, not proof that every such interference was eliminated.
For the test, the reviewers supplied an RTSP stream from the Hikvision camera. They configured two virtual lines, assigned an entry direction to one and an exit direction to the other, and set expected object sizes. A server solution still required camera positioning and calibration; it did not make scene geometry irrelevant.
The corridor narrative reported 98.15% for entry and 95.65% for exit. In the elevator lobby, it reported 96.77% entry and 86.36% exit. The source cited a vendor claim of at least 95% accuracy, but its own difficult-scene exit result was below that figure. It would therefore be misleading to describe the test as confirming a universal 95% minimum.
Preserved results and their limits
| Solution | Corridor entry | Corridor exit | Elevator-lobby result in the narrative |
|---|---|---|---|
| Megacount VideoCount 3D ASSIS | 98.1% | 95.65% | Unsatisfactory; no separate exact narrative values |
| CountBOX STD-1 | 98.8% | 97.8% | Unsatisfactory; no separate exact narrative values |
| Hikvision DS-2CD4A26FWD-IZHS | Severe overcounting | Severe overcounting | 88% reported without a directional breakdown in the text |
| Smartec STC-IPMX3193A | 96.4% | 95.2% | Unsatisfactory; no separate exact narrative values |
| Acumen AiP-P25V-45Y2B | 92.6% | 91.3% | Entry 83.1%; exit 60% |
| Synesis Cassiopeia demo | 98.15% | 95.65% | Entry 96.77%; exit 86.36% |
These are the percentages reported in the archived prose. The original charts used a different presentation: device count relative to the manual reference, with 100% representing an exact match, values below 100% undercounting, and values above 100% overcounting. That ratio is not the same as an accuracy score.
The charts showed substantial overcounting for several difficult-scene results and severe Hikvision overcounting in the corridor. Exact raw counts, sample sizes, confidence intervals, and a consistently defined accuracy formula were not supplied. The table therefore preserves the narrative without inventing precise values from unlabelled bar heights or combining incompatible metrics.
Feature differences in the reviewed configurations
| Solution | Object-size adjustment | Tracking overlay | Report export | FTP report transfer |
|---|---|---|---|---|
| CountBOX | Yes | Not listed | Yes | Yes |
| Megacount | Yes | Yes | Yes | Yes |
| Hikvision | Not in tested interface | Not listed | Yes | Not listed |
| Acumen | Not listed | Not listed | Yes | Not listed |
| Smartec | Yes | Yes | Yes | Not listed |
| Cassiopeia demo | Yes | Yes | Yes | Yes |
The source matrix described downward and angled placement as available across the compared approaches, but this did not establish successful operation at arbitrary angles. Historical prices and supplier promotions have been omitted because they do not describe a current installation budget. FTP support is likewise a historical feature description, not a recommendation to expose an unprotected transfer service.
Apply the lessons to a new installation
Define the counting task before selecting hardware: total crossings, direction-specific visits, occupancy, traffic profiles, or conversion analysis. Decide how staff, repeated crossings, groups, deliveries, and people waiting near the entrance should be treated. These definitions affect both configuration and interpretation.
Give the detector enough visible approach and departure space, follow its mounting guidance, and avoid placing the count line where people routinely stop or turn. Where the physical layout cannot be improved, evaluate products in that exact scene and include specialist calibration when it is part of the intended deployment.
Validate with manual reference counts over representative quiet and busy periods. Report entries and exits separately and record the actual error definition. Recheck after changes to camera position, furniture, lighting, software, or customer flow. Do not extend a high percentage from a favorable short test into a permanent guarantee.
Counts and movement patterns can support better business decisions when they remain trustworthy. Compare compatible security cameras and video systems, then assess analytics licenses, data export, calibration, and operating support as part of the complete solution.
Related products from Safsale
Continue from Academy guidance to live equipment and components available on Safsale.
