Seven weeks of Checkers Sixty60 delivery-bike data

Date prepared: 1 July 2026.

Analysis period: 10 May 2026 to 30 June 2026.

Purpose: to provide a clear evidence record of repeated Checkers Sixty60 delivery-bike passes through a residential route, based on resident-held camera detection data. This page records observations and estimates. It does not make a legal finding.

The central issue shown by this data is scale. This is not an occasional local delivery pattern. On the trusted full recording days, one residential stop-street point detected a minimum average of 136 Checkers Sixty60 delivery-bike passes between 08:00 and 18:00 alone. That detected average is not the full daily total, because it excludes missed fast stop-street passes, additional bikes in simultaneous groups, and the 18:00 to 20:00 delivery period where low light makes detection unreliable. After known undercount factors and the unobserved evening delivery window are accounted for, the estimated average is about 202 passes per full delivery day.

What was measured

A fixed camera observed a stop-street area on an otherwise quiet residential road. Residents report that the road is being used by Checkers Sixty60 delivery motorbikes as a cut-through to reach nearby residential estates.

For this analysis, one pass means one Checkers Sixty60 delivery motorbike crossing the monitored intersection once. Other motorcycles, scooters, and general traffic are detected by the broader system but filtered out of the Checkers-positive counts reported here.

A significant limitation is speed through the stop-street area. When riders pass very quickly, the camera can capture only a blurred crop or no usable crop at all. Those events can fail detection or classification, even when residents hear or see the bike pass.

The published detected count should be read as the number of confirmed passes the system caught, not as the full number of bikes that used the route. When two or more Checkers Sixty60 bikes pass together, the public count can miss or collapse the additional bike or bikes. Residents observe this kind of simultaneous pass a few times a day.

The public article deliberately avoids exact location details and identifying rider information. The purpose is to show frequency, timing, and residential impact patterns, not to identify private individuals.

How the data was gathered

In plain language, a camera points at the intersection. Software watches the live video feed and detects motorbikes moving through a defined count zone. When a tracked motorbike enters that zone, the system saves a cropped image and logs the event with a timestamp.

A separately trained image classifier then sorts saved crops into Checkers delivery bikes and non-Checkers bikes. Detections of the same Checkers-positive bike within three seconds are merged into one pass, so repeated frames or track fragments do not inflate the count.

The system depends on a usable crop image. If a rider passes through the stop-street area too quickly, motion blur can make the crop unusable or prevent a reliable trigger. This is a known undercount source, not a reason to treat the detected totals as complete.

• Live RTSP feed from a fixed camera, processed at 15 frames per second.
• Object detector: YOLO model in recall-first mode with a low confidence threshold of 0.10.
• Tracking: ByteTrack multi-object tracking to follow bikes across frames.
• Classifier: purpose-trained binary YOLO classification model using 224px input crops.
• Classifier training set: about 2,600 hand-labelled crops.
• De-duplication: Checkers-positive detections within three seconds are collapsed into one pass.
• Storage: local SQLite database with timestamp, confidence score, and representative photo per event.

Coverage and limitations

Recording coverage was uneven. The detector runs on a local laptop that is sometimes needed for other work, and the process can occasionally hang silently. For that reason, the headline statistics do not use every calendar day in the 10 May to 30 June window.

• 52 calendar days were reviewed in the broader analysis window.
• 45 days had some recording, 31 were trusted full days, and 14 were partial or invalid for full-day analysis.
• 7 days had no recording.
• A trusted full day means each hour from 08:00 to 18:00 had usable detection coverage and at least one detected pass.
• No Thursday had full-day coverage, so Thursday is absent from the weekday comparison.

The camera is also a daylight tool. In winter, reliable detection ends around 18:00 even though Checkers Sixty60 deliveries continue later. Low light after 18:00 makes bike crops harder to capture and classify, which causes missed detections. None of the detected charts or detected tables include the 18:00 to 20:00 delivery period. That evening window is modelled separately and labelled as estimated, not observed.

Fast stop-street passes are another important limitation. Residents regularly observe bikes passing too quickly for the camera to capture an unblurred image. Those passes can be missed even during otherwise trusted recording windows.

How the estimates were calculated

The detected figure is the foundation: a conservative minimum of 4,216 Checkers Sixty60 passes from 08:00 to 18:00 across 31 trusted full days. That figure is a direct database count after classification and de-duplication. It is not a full route total.

The true daytime estimate adjusts for the observed missed-detection rate. Residents observed that the detector catches roughly 75% to 80% of passes in this setting, largely because many riders pass through the stop-street area too quickly for the camera to capture a clear, unblurred crop. The detected count is therefore multiplied by 1.25 to 1.33. That produces an estimated true daytime total of about 5,300 to 5,600 passes.

That estimate should still be read conservatively. When multiple Checkers Sixty60 bikes pass together, the public count can miss or collapse the additional bike or bikes, and those simultaneous-bike undercounts are not separately itemised in the tables.

The full delivery-window estimate then adds a modelled winter evening period. Because low light makes the camera unreliable after 18:00 and causes missed detections, 18:00 to 19:00 is modelled as 65% of the same day's 17:00 hour, and 19:00 to 20:00 is modelled as 35% of the same day's 17:00 hour. These evening assumptions are estimates and should be read as such.

What this evidence supports

This dataset supports a narrow factual conclusion: even the conservative minimum detected count shows that repeated Checkers Sixty60 delivery-bike passes through this residential point are frequent, sustained, and concentrated in times when residents would normally expect residential quiet, including weekends and the late afternoon.

The data does not by itself decide any legal question. It does, however, give the City of Cape Town, Shoprite/Checkers, and affected residents a concrete basis for evaluating whether the current routing and vehicle pattern is reasonable for a residential street.

The practical mitigation request remains measurable: reduce unnecessary residential shortcut passes, keep through-delivery traffic on main roads where possible, move high-frequency residential delivery work to quieter vehicles, and provide accountable follow-up that can be checked against before-and-after data.