{"id":"af1f87c4-791e-4cd5-8302-1c072715fdf9","arxiv_id":"1908.08892","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A camera measures distance to known LED lights by the size of their image on the sensor, then triangulates the camera's position, using this for indoor smartphone and outdoor vehicle localization.","lead":"This thesis describes two ways to find a device's location using LED lights and a camera: one for smartphones indoors, one for cars on the road. It combines the light signals with image-based distance measurement, called photogrammetry, so the same lights used for illumination can also act as positioning beacons.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 10 cm indoor localization claim rests on Eq. 18's perpendicular-plane assumption, but the actual trilateration geometry views most LEDs off-axis; perspective foreshortening is acknowledged in Section 6.5.1 and never corrected, so the accuracy is not established for the proposed setup.","rationale":"The reader's weakest assumption identifies the same load-bearing concern: Eq. 18 assumes a perpendicular LED plane, while the proposed trilateration necessarily views LEDs at different off-axis angles. This is the most consequential gap because every distance feeding Equations 26-34 inherits the perspective foreshortening bias, and Section 6.5.1 names the distortion but does not integrate a correction into the algorithm or into the simulation results supporting the 10 cm claim. The manual example in Figure 17 is also illustrative rather than quantitative: the three LEDs are collinear, the camera is under the line, and no error analysis is given. I agree with the CONDITIONAL verdict because the concern is addressable: a synthetic rendering test or a lab experiment with off-axis LEDs at realistic room scale would either confirm the bias or show that the proposed geometry keeps errors below 10 cm. No code, data, or full experiment is provided, but the issue is concrete and testable, so REJECT is not forced; rather, the current conditional status is the appropriate outcome until the off-axis test is run.","tokens_in":950,"tokens_out":881,"duration_ms":50968,"concrete_test":"Run a synthetic pinhole-camera test of Section 4.3.1: place three 10x10 cm ceiling LEDs in a triangle (e.g., 300 cm ceiling height, 300 cm spacing), render their projections with the stated focal length (16-25 mm) and pixel size (2.5-4 um) using true perspective projection, count the projected pixels, feed the counts through Eq. 18, and solve the trilateration exactly as in Equations 26-34. Compare the recovered smartphone position with ground truth for several camera positions, and also compute the bias at off-axis angles 0, 15, 30, and 45 degrees for a single LED. If the error exceeds 10 cm while the same pipeline with LED normals forced parallel to the optical axis gives under 10 cm, the claimed accuracy is an artifact of the perpendicular-plane assumption rather than a property of the proposed system.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.3.1 derives Eq. 18, d = f * sqrt(A_LED / (rho^2 * eta_i)), from the lens-magnification relation M = f/d, which assumes the LED plane is perpendicular to the camera's optical axis. In the indoor scheme, at least three ceiling-mounted LEDs are in view (Figure 7), and except for an LED near the image center each is viewed at a significant off-axis angle. For a horizontal LED panel of area A viewed at angle theta from its normal, the projected image area is approximately A*cos(theta)*f^2/d^2, not A*f^2/d^2, so Eq. 18 returns d/sqrt(cos(theta)). At theta = 20 degrees this is about 3.1% high, or roughly 9 cm at a 3 m distance; at theta = 35 degrees it is about 10% high. Trilateration in Section 4.3.1 then combines several such biased distances when solving Eq. 26-34, so the error propagates directly into the claimed '10 cm level localization resolution' of Section 8.1. Section 6.5.1 explicitly acknowledges 'perspective distortion caused due to the relative position between the object and the camera' and mentions a back-propagation correction, but no such correction appears in the localization algorithm or in the simulation results of Figures 30-33. The bench experiment in Figures 40-41 only measures a centered LED at short range and does not test off-axis geometry, so it does not validate the accuracy claim either.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript, an M.S. thesis posted on arXiv, proposes localization schemes for indoor smartphones and outdoor vehicles by combining optical camera communication (OCC) and photogrammetry. Indoor positioning uses LED-ID reception to identify ceiling lights and measures the camera-to-LED distances from the projected image area on the image sensor (Eq. 18), followed by trilateration and Kalman filtering. Vehicle positioning uses rear lights of forwarding vehicles and street lamps as optical transmitters, with photogrammetric distance estimation and a set of asserted coordinate relations (Eqs. 48–53). The central claims are a 10 cm level indoor localization resolution and a vehicle positioning scheme that achieves roughly 90% accuracy under certain street-light spacings, supported by simulation plots and a short-range benchtop distance experiment.","tokens_in":44133,"tokens_out":2721,"duration_ms":29759,"significance":"If the claims hold, the work offers a inexpensive way to reuse LED lighting infrastructure and smartphone/vehicle cameras for simultaneous communication and localization, which is relevant for IoT, LBS, and vehicular safety. The derivation of Eq. (18) from the lens equation is a transparent, parameter-free photogrammetric distance formula, and the simulated BER/SNIR comparisons in Figs. 29 and 34 provide a useful baseline. However, the significance is limited by the absence of any off-axis error analysis, the lack of a derivation for the vehicle localization equations, and the thin experimental validation, so the central accuracy claims are not yet established.","major_comments":[{"comment":"The distance formula d = f sqrt(A_LED) / (rho sqrt(eta_i)) assumes the LED plane is perpendicular to the optical axis (magnification M = f/d). In the indoor trilateration geometry of §4.2, LEDs at the edge of the camera FOV are viewed obliquely, so the projected area is foreshortened by approximately cos(theta) and Eq. (18) returns a biased distance. Section 6.5.1 explicitly acknowledges 'perspective distortion caused due to the relative position between the object and the camera' and mentions a back-propagation correction, but no such correction is incorporated into the localization algorithm of §4.3.1 or into the simulation results of Figs. 30–33. This uncorrected bias propagates into every distance measurement and directly into the claimed 10 cm resolution, so the central accuracy claim is not supported without either a corrected model or a quantitative error analysis for off-axis LEDs.","section":"§3.3.1, Eq. (18); §6.5.1"},{"comment":"The vehicle localization equations are asserted without derivation. Eq. (48) states that the total position shift of the host vehicle is a function of h_j, n_IS_SL, V_HV, c_j, d_SL-SL, and delta-t, but the manuscript never derives how these quantities combine, how pixel area and velocity are weighted, or what coordinate frame the result is expressed in. Similarly, Eqs. (50)–(53) relate angular position, pixel area, and flat displacement without a geometric or kinematic derivation. Because these equations are load-bearing for the claimed vehicle positioning accuracy (Figs. 38–39), the manuscript needs a step-by-step derivation or a clear citation to a published derivation.","section":"§5.3.1, Eqs. (48)–(53)"},{"comment":"The simulation results that support the 10 cm indoor localization claim are presented as plots without error bars, without a description of the noise model, number of Monte Carlo runs, or parameter distributions, and without a quantitative metric such as RMSE or a CDF. The text says 'within 9 to 10 cm, the position of the smartphone is estimated' (near Fig. 30), but the figures do not show the error distribution or the dependence on LED geometry. Without this methodology, the central accuracy claim is not reproducible or verifiable.","section":"§6.1, Figs. 30–33"},{"comment":"The experimental validation measures only a single, centered LED at distances of roughly 0.6 m and 2.5 m (and up to ~0.65 m in Fig. 41). It does not test off-axis LEDs, multi-LED trilateration, or the smartphone-localization scenario of §4, so it cannot validate the 10 cm indoor localization claim. The experiment also reports error within 1% for distance measurement, which would be useful if it were extended to the actual off-axis geometry used in the proposed scheme.","section":"§6.3, Figs. 40–41"}],"minor_comments":[{"comment":"There are typos such as 'e-commence sector' and 'a new era has written' in the abstract; the manuscript would benefit from a careful language edit.","section":"Abstract and Chapter 1"},{"comment":"Several equations contain garbled or missing symbols, particularly Eq. (7) and the notation list (e.g., the Lambertian index and cosine terms), which makes them difficult to verify. The authors should ensure all symbols are rendered correctly.","section":"§3.2.3, Eq. (7)"},{"comment":"The transition from Eq. (26) to the matrix form in Eq. (27) is not fully explained, and the notation for the homogeneous solution in Eqs. (29)–(32) is confusing; a brief explanation of how the particular solution and the null space are obtained would improve clarity.","section":"§4.3.1, Eqs. (27)–(34)"},{"comment":"The derivation of c and h in the vehicle scenario is compressed; labeling the geometric variables in Fig. 23 consistently with the equations would help the reader follow the Pythagorean steps.","section":"§5.3.1, Eq. (44)–(47)"},{"comment":"The text states that the maximum localization accuracy is approximately 90% when the distance between street lights is 40 m, but the y-axis label reads 'Vehicle positioning accuracy (%)' and the curve appears to vary with the number of street lights in FOV; please clarify the definition of accuracy and report the error bars or confidence intervals.","section":"§6.2, Fig. 39"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is an M.S. thesis rather than a journal-style paper, and the presentation is often thesis-like (acknowledgments, list of publications, lack of a peer-reviewed format). The heavy reliance on the author's own published papers for the core schemes (e.g., refs. [102], [106], [110]) is worth noting; while not improper, it means the novelty relative to those papers should be explicitly stated. The main technical issue is the uncorrected off-axis perspective distortion, which should be treated as a correctness risk in the revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a master's thesis that packages the author's own 2018 journal papers on LED/camera localization. The distance formula is derived correctly, but the 10 cm accuracy claim is not supported for the geometry the scheme actually uses.\n\nWhat's good: Eq. 18 is a clean derivation from the lens equation, and the bench experiment in Fig. 41 shows under 1% distance error for a centered LED at short range. The thesis is also candid about practical challenges (distortion, ambient light), and the Kalman filter description is standard but accurate.\n\nThe soft spot is load-bearing. Eq. 18 assumes the LED plane is perpendicular to the optical axis. In the indoor scheme, three or more ceiling LEDs are in view, and all but the one near the image center are seen obliquely. That foreshortens the projected area, so Eq. 18 returns d/sqrt(cos(theta)). At 20 degrees off-axis that is about 3% high; at 35 degrees, 10% high. Those biased distances go straight into the trilateration solve. Section 6.5.1 acknowledges perspective distortion and mentions a back-propagation correction, but no correction appears in the localization algorithm or in the simulation figures. The simulation plots (Figs. 30-33) have no error bars and no methodology; the 9-10 cm resolution is one curve. The vehicle equations (48)-(53) are asserted without derivation and there is no experimental validation. No code or data is provided.\n\nAlso, the two schemes were already published by the same author in 2018 (refs [102] and [106]), so as an arXiv preprint this is a thesis compilation rather than a new result. The self-citation is fair — the journal papers are the primary sources — but it means a journal editor should treat this as a derivative manuscript.\n\nWho gets value? A newcomer to VLP who wants one document walking through the OCC-plus-photogrammetry idea, or a student writing a thesis in the area. A researcher in visible light positioning won't find a new technique, and the headline accuracy is unproven for the proposed setup.\n\nMy recommendation: if this came in as a journal submission, the duplication alone justifies a desk reject. But the derivation is checkable and the perspective issue is fixable, so I would rather send it to review with a mandatory demand for a foreshortening correction and real experimental data than bounce it without referee input. The central idea may survive the fix.","headline":"A competent thesis compilation of already-published OCC/photogrammetry schemes; the 10 cm accuracy claim is unproven because the distance formula ignores off-axis foreshortening.","tokens_in":44657,"tokens_out":2744,"would_cite":false,"duration_ms":28362,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A camera can localize a phone indoors to about 10 cm using only ceiling LED lights and their projected image sizes.","keywords":["optical camera communication","photogrammetry","indoor localization","smartphone localization","vehicle localization","image sensor","trilateration","Kalman filter"],"falsifier":"Mount a camera at a surveyed position beneath three or more ceiling LEDs, read each LED-ID, measure each projected area on the image sensor, apply $d = f\\sqrt{A_{\\mathrm{LED}}}/(\\rho\\sqrt{\\eta_i})$ to compute the three distances, and compare the trilaterated position with the surveyed position while moving the camera toward the edge of the field of view; if the error grows systematically with the offset angle and exceeds 10 cm, the central accuracy claim fails.","tokens_in":43611,"feed_emoji":"📷","tokens_out":7548,"duration_ms":71833,"temperature":0.7,"pith_summary":"This thesis proposes using light that is already in the environment as a positioning infrastructure. A smartphone camera reads the identity of ceiling LED lamps from modulated light, measures how far away each lamp is from how large it appears on the image sensor, and combines three or more such distances to fix its own position; simulation results put that position estimate at about 10 cm. The same combination is applied on the road, where a host vehicle's camera reads street-light IDs and the taillight IDs of vehicles ahead and uses projected image areas to estimate distance. The point of the scheme is that it needs almost nothing new: ordinary LED lighting, a camera, and a server that knows the LED coordinates. If the claims hold, indoor navigation and vehicle proximity awareness could be delivered with little modification of infrastructure.","feed_headline":"Phone camera + ceiling LEDs pin position to ~10 cm","feed_subtitle":"The same projected-area trick reads street-light and taillight distances, giving vehicles a positioning fix from existing lights.","key_machinery":"The load-bearing identity is the photogrammetric distance formula $d = f\\sqrt{A_{\\mathrm{LED}}}/(\\rho\\sqrt{\\eta_i})$ (Equation 18), which converts the measured pixel area $\\eta_i$ of an LED's image on the sensor into a distance, using the real LED area $A_{\\mathrm{LED}}$, focal length $f$, and pixel size $\\rho$. The formula follows from the pinhole lens-magnification relation $M = f/d$, and it assumes the LED plane is perpendicular to the optical axis. These distances feed a standard trilateration solve to produce coordinates, and a Kalman filter recursively predicts the next position to absorb motion error. For vehicles, the same projected-area distance measurement is applied to street lights and taillights, with S2-PSK modulation carrying LED-IDs and marking the regions of interest.","core_discovery":"The thesis claims that a single camera can act as both a communication receiver and a rangefinder so that the device carrying the camera can locate itself. Indoors, each ceiling LED broadcasts its own coordinates as a modulated LED-ID; the camera decodes the IDs, measures the pixel area each LED occupies on the image sensor, and converts that area into a distance with the photogrammetric relation. Three or more such distances plus the known LED coordinates give the camera position by trilateration. Outdoors, the same principle measures the host vehicle's distance to street lights and to the taillights of forwarding vehicles, giving relative positions while vehicles move; a Kalman filter smooths the indoor estimate over time. Simulations reported in the thesis give about 10 cm resolution for the indoor case and roughly 90 percent vehicle positioning accuracy when four street lights about 40 m apart are in view.","pith_inferences":["A direct extension would be to replace the perpendicular-plane distance formula with a perspective-aware version that uses the decoded LED coordinates to estimate each LED's viewing angle, then inverts the oblique projection before computing distance; if the systematic bias shrinks, the 10 cm claim would hold across a wider field of view.","The outdoor scheme points toward cooperative positioning: if each vehicle broadcasts its own ID, a chain of vehicles could propagate position estimates without needing every vehicle to see a street light directly.","A reverse use is plausible: fixed roadside cameras that observe a vehicle's taillights could apply the same projected-area relation to localize the vehicle from infrastructure, which would be useful in intersections and tunnels."],"forward_implications":["If the claimed accuracy holds, indoor location-based services can run on the existing ceiling LED infrastructure plus the phone's own camera, without new radio beacons.","Because each LED has a unique ID, the localization is inherently self-identifying: the phone knows which lamp it is seeing, so the measurement can be tied to a coordinate server.","The same camera link that serves localization also carries data, so a vehicle can receive an ID from a taillight and estimate its distance to that vehicle at the same time.","Kalman-filter tracking turns the static distance measurements into a smooth position estimate, so a moving phone or vehicle can be followed between frames.","The simulation results suggest that for vehicles, the practical limit of the scheme appears when street lights are spaced near 40 m and when at least four lights remain in view; beyond that spacing the positioning accuracy falls."],"supporting_citations":[{"why":"It supplies the constant-tau distance relation used to read distance from pixel count for a given camera and LED fixture.","marker":"[19]"},{"why":"It provides the pinhole camera model that maps world coordinates to image coordinates, the basis of the distance formula.","marker":"[101]"},{"why":"It is the prior indoor localization scheme this thesis builds on to combine LED-ID reception with photogrammetry.","marker":"[102]"},{"why":"It provides the trilateration solution used to compute camera coordinates from three or more distances.","marker":"[103]"},{"why":"It supplies the Kalman filter recursions used to track a moving position and reduce estimation error.","marker":"[104]"},{"why":"It is the vehicle localization scheme that combines OCC and photogrammetry for positioning a host vehicle and forwarding vehicles.","marker":"[106]"},{"why":"It provides the S2-PSK modulation and region-of-interest signaling used to receive FV-IDs and SL-IDs in the vehicle scenario.","marker":"[110]"}],"fun_headline_variants":["LEDs + camera give 10-cm positioning via photogrammetry","Camera reads LED IDs to locate within 10 cm","OCC + photogrammetry: 10-cm fix from existing lights","Existing LEDs turn camera into a 10-cm localizer","Smartphone camera + LEDs = 10-cm location fix"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the measured pixel area of an LED acts as a faithful distance cue, treating every LED as if it faced the camera head-on; LEDs seen near the edge of the field of view are actually viewed at an angle, and that perspective shortening is not corrected in the localization calculation.","fun_headline_variants_meta":{"raw":{"variants":["LEDs + camera give 10-cm positioning via photogrammetry","Camera reads LED IDs to locate within 10 cm","OCC + photogrammetry: 10-cm fix from existing lights","Existing LEDs turn camera into a 10-cm localizer","Smartphone camera + LEDs = 10-cm location fix"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000604,"raw_usage":{"total_tokens":2818,"prompt_tokens":942,"completion_tokens":1876,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":558,"completion_tokens_details":{"reasoning_tokens":1787}},"tokens_in":558,"tokens_out":1876,"duration_ms":12594,"temperature":1.0,"reasoning_tokens":1787,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:26:30.762032+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Mount a camera at a surveyed position beneath three or more ceiling LEDs, read each LED-ID, measure each projected area on the image sensor, apply $d = f\\sqrt{A_{\\mathrm{LED}}}/(\\rho\\sqrt{\\eta_i})$ to compute the three distances, and compare the trilaterated position with the surveyed position while moving the camera toward the edge of the field of view; if the error grows systematically with the offset angle and exceeds 10 cm, the central accuracy claim fails.","supporting_citations":[{"cited_title":"Hartley and A","cited_arxiv_id":null,"evidence_quote":"It provides the pinhole camera model that maps world coordinates to image coordinates, the basis of the distance formula."},{"cited_title":"A novel indoor mobile localization system based on optical camera communication,","cited_arxiv_id":null,"evidence_quote":"It is the prior indoor localization scheme this thesis builds on to combine LED-ID reception with photogrammetry."},{"cited_title":"Revisiting trilateration for robot localization,","cited_arxiv_id":null,"evidence_quote":"It provides the trilateration solution used to compute camera coordinates from three or more distances."},{"cited_title":"Ristic, S","cited_arxiv_id":null,"evidence_quote":"It supplies the Kalman filter recursions used to track a moving position and reduce estimation error."},{"cited_title":"A new vehicle localization scheme based on combined optical camera communication and photogrammetry,","cited_arxiv_id":null,"evidence_quote":"It is the vehicle localization scheme that combines OCC and photogrammetry for positioning a host vehicle and forwarding vehicles."},{"cited_title":"Region -of-interest signaling vehicular system using optical camera communications,","cited_arxiv_id":null,"evidence_quote":"It provides the S2-PSK modulation and region-of-interest signaling used to receive FV-IDs and SL-IDs in the vehicle scenario."}],"review_version":1}