Asphalt-edge notches: vegetation, real damage, or detector artifacts?

persp_cam005_s242_zoom_depth.png is decisive: a continuous yellow (~0.6 m) guardrail line marks the true outboard limit, and between it and the red-notch crosshair the surface is grey (road-plane height) speckled with orange 0.2-0.4 m returns, i.e. low weeds/grass sitting on top of pavement that continues outboard of the detected edge. The RGB zooms (cam005/cam004) confirm a dark olive band of grass/weeds along the guardrail encroaching over the light-grey shoulder exactly where the red edge jogs inboard of the cyan trend, and topdown_zrange shows the tall/bright band bulging inward at the same station. Only rasters+perspective are available for A3 (no point-cloud renders), hence moderate confidence.

topdown_pc_dz shows a raised band (dz 0.2-0.8 m) hugging the verge whose tongue pushes onto the pavement exactly at the notch apex, and the detected red edge bends inward to wrap that tongue; pc_stats confirms it is vegetation-like (bite_low_green_med 0.046 vs ref 0.0067, low-point density 714 vs 2467). xsec_along_NN at +1.0 and +2.0 m shows a cyan canopy cluster ~0.7 m above the surface just outboard of the red dashed edge with the flat surface line continuing underneath out to about -1.5/-2.5 m, and topdown_pc_low still has pavement-level (non-magenta) cells 1-1.5 m outboard under that band. Secondary dropout: the large magenta/black wedge in topdown_pc_low, topdown_intensity and the perspective views is the occlusion shadow behind the same vegetation, which removes the returns the detector would have needed further out.

topdown_intensity shows a smooth, uniform bright pavement return continuing ~50 px (px-sampled mean ~160-170 at x=270-300 on every notch row) outboard of the detected edge, which drops from x~307 above the notch to x~252 through it — i.e. the pavement surface does not end at the notch. topdown_zrange shows the tall/rough boundary sitting at x~253-270 exactly where the red line dives in, so the detector followed low vegetation encroaching over the asphalt rather than the pavement edge; persp_cam001/002_zoom_depth confirm a 0.2-0.6 m (tan/yellow) grass verge band immediately outboard with road-plane grey continuing under it, and no green/white tall canopy or data hole.

xsec_along_NN (all three panels) shows the flat asphalt surface ending at lateral ~0 with the point cloud rising immediately into a 1+ m vegetated bank — no flat pavement-level returns continue outboard of the detected edge. topdown_pc_low confirms it: with the canopy stripped, the band outboard of the red edge is olive ground-vegetation and then magenta, never grey pavement, so the pavement genuinely stops at the hedge. The 0.49 m "notch" is measured against a smooth trend that is itself pulled outboard at along-edge +1.5 to +2.5, where the edge bulges into a locally flat low-vegetation patch (visible as the dark/purple intrusion inside the yellow band in topdown_zrange) — i.e. the candidate sits on the true edge and the neighbours are wrong.

No vegetation or object is involved: topdown_pc_dz is flat (~0 m above the asphalt plane) everywhere near the edge, pc_low keeps 97% of points and is essentially identical to topdown_pc_rgb, and xsec shows no canopy above the surface and no data hole in the notch. The in-notch cross-sections (xsec_along_NN, +0.0/-1.0/+1.0 m) show the detected edge sitting exactly on a genuine break of slope, with a brighter, steeply falling gravel shoulder/drainage swale outboard — pavement really stops there. The apparent notch is largely because the OUTSIDE-notch sections (-3.9 m) have the edge placed ~2.5 m too far outboard, down in the ditch bottom, while the true asphalt step there is at +1.15 m; so the geometry is real but the surrounding edge is the misplaced part (topdown_intensity_wide also shows blocky staircase steps along this stretch).

All three panels of xsec_along_NN.png show a sharp, repeatable material break at lateral x=0: dark low-intensity asphalt inboard, a ~4-5 cm step/berm and higher-intensity (green) flat gravel/soil shoulder outboard — i.e. the asphalt genuinely ends where the notch line sits, and the notch apex is actually the correct edge while the neighbouring polyline at x≈-0.7 overshoots onto the shoulder. topdown_pc_low.png and topdown_pc_dz.png show that outboard band fully populated with dz≈0 (no canopy stripped, no second surface beneath), and topdown_zrange.png is flat/dark over the notch, with the only tall returns being the continuous guardrail band ~0.8-1.2 m outboard, which runs the whole 10 m and therefore cannot explain a local notch. No dropout (cells have points), no local object; the sparse elevated spray at x=-1.1..-1.6 m in the centre slab is vegetation at the guardrail, too far outboard to drive a 0.5 m notch.

Both perspective zooms (persp_cam003/004_s155_zoom.png) show a roadworks zone: a "Krombacher" billboard hoarding, green temporary fence panels and mesh barrier standing on the paved left shoulder, with no vegetation anywhere near the crosshair (trees are far off-scene). The depth views (persp_cam00{3,4}_s155_zoom_depth.png) confirm the candidate crosshair sits on grey road-plane pavement while the detected polyline (magenta/cyan dots) has been dragged outboard onto the tall billboard/fence line and then hooks back, producing the notch; grey pavement clearly continues under and behind the barrier. topdown_zrange.png shows a bright tall-object band exactly along that edge stretch, consistent with a man-made barrier rather than canopy.

The detected edge leaves the true grass boundary at y=-6.3 and jumps ~6 m into the carriageway (topdown_pc_low / topdown_pc_dz), yet the pavement there is fully intact: xsec_along_NN shows one continuous flat asphalt surface from x=-6 to +3 with no break, and dz is uniformly ~0 across the whole notch, so this is neither a real notch, vegetation nor dropout. The trigger is a small raised yellow object lying on the road at along-edge -1..+0.5 (bright V in topdown_zrange, yellow smear at the crosshair in persp_cam000/001_zoom, ~15 cm high yellow specks above the surface in the central xsec panel), whose high-zrange cells the detector latched onto together with the dark longitudinal seam at y=0 in topdown_pc_int.

Both perspective zooms (persp_cam000_zoom.png, persp_cam004_s261_zoom.png) put the red notch crosshair squarely on continuous asphalt several metres inside the roadway — the notch tip is ~7.4 m off the expected edge (persp_meta.json) — and the occluder is not vegetation but a road overpass: cam000_zoom shows dark horizontal bands of bridge-deck points laid over the carriageway, which persp_cam000_zoom_depth.png renders as saturated white (far above the fitted road plane), and topdown_zrange.png shows the notch tip sitting exactly on the boundary of the large high-z-range (yellow) bridge footprint. Pavement is plainly visible under the structure in RGB, so this is an overhead-object artifact, aggravated by a band of grey no-data / very-low-intensity cells outboard of the edge in topdown_zrange.png and topdown_intensity.png. Not vegetation and not a real edge; A3 has no point-cloud renders so confidence is held moderate.

In topdown_intensity/topdown_zrange the notch is a razor-straight, near-vertical no-data cut (black in BOTH rasters, ~10 m deep, running the full crop height) and the detected red edge simply traces that boundary — a scan/tile coverage boundary, not a physical feature. The zrange cells immediately adjacent to the hole are flat grey with no bright (tall) rim, so no canopy or object is occluding it; topdown_intensity_wide shows the surrounding shoulder texture continuing uniformly right up to the cut. Perspective persp_cam003/004_zoom_depth show the crosshair sitting on flat road-plane-height ground with no green (vegetation) or raised blob nearby, confirming no vegetation/object cause.

In topdown_intensity/_wide the red polyline kinks exactly on the corner of a black zero-return wedge in the raster — the notch traces the scan-coverage hole, not any physical feature, and topdown_zrange is uniformly flat dark (no tall stuff anywhere near the candidate; the only bright vegetation is far off in the bottom-right corner). The cam003/cam004 depth zooms confirm continuous grey height=0 asphalt on both sides of and well outboard of the cyan edge polyline at the crosshair, with the yellow/green canopy only in the far background. So the detector fell into a data hole on open pavement.

xsec_along_NN shows the flat asphalt surface running unbroken through all three in-notch slabs out to lateral -1.1 m with nothing above it except a continuous guardrail/kerb feature that is present in the OUTSIDE slabs too, so no canopy or object sits over the bitten area; topdown_pc_dz confirms the whole notch is uniform dz~0 pavement plane. topdown_pc_rgb/_low show the bite filled with normal asphalt-intensity cells but peppered with an unusually dense magenta (zero-point) speckle, and pc_stats gives bite_low_density 5598 vs ref 15285 pts/m2 at matching intensity (27218 vs 26170) — i.e. real pavement with ~1/3 the return density. The detector fell into a thin-return/scan-shadow hole, not a physical edge feature.

The "notch" is a 0.75 m wide, 6 m deep needle-thin spike of the detected polyline shooting across intact asphalt: xsec_along_NN at +0.0 m puts the red dashed edge at lateral -2.9 m in the middle of a perfectly continuous flat surface, while the true pavement break (step down to soil) sits at -6.0 m exactly where the ±1 m sections place it. topdown_pc_low and topdown_pc_dz show uniform pavement-level returns and dz ~0 over the whole spike — no canopy, no magenta dropout, no raised blob — and topdown_intensity shows the spike tracing a thin dark crack/joint line in the asphalt. So this is a detector false positive following an intensity seam, not any of vegetation/object/dropout/real irregularity; none of the physical categories fits, hence unclear.

None of the physical causes apply: xsec_along_NN shows one continuous, unbroken flat asphalt profile across all five slabs with no step, no canopy points above and no gap at the detected edge, and topdown_pc_low/topdown_pc_rgb show fully populated grey asphalt with lane markings on both sides of the red polyline (zero magenta), while topdown_pc_dz is uniformly ~0 m (no vegetation or object relief). persp_cam007_ctx places the candidate mid-carriageway in a motorway gore/exit area, with the true grass boundary several metres further outboard, and persp_cam008_zoom shows the polyline hugging a painted line / shadow boundary rather than any physical edge. This is a detector artifact on continuous pavement (intensity/marking-driven edge wander), so it fits neither vegetation_overhang, real_notch, object_on_edge, nor dropout.

xsec_along_NN (+0.0 m panel) shows the asphalt surface running unbroken from lateral -4 m out to +3 m with the detected edge planted at 0 m, i.e. ~4 m inside the true pavement boundary; topdown_pc_rgb/_low and persp_cam000_zoom confirm the notch sits on open grey asphalt right at the painted white edge line, with the real grass/vegetation boundary a full 4 m outboard. No canopy, object, or magenta dropout in the notch area, so none of the physical causes apply — this is a detector artifact snapping the edge onto the road marking, hence unclear rather than vegetation_overhang or real_notch.