Patent Pending Sensor Sorting

Sensor Feeds

Sensors see everything, including the junk. I sort the junk out.

Sensor Feeds radar hexagon

Sensor Feeds

Lidar, radar, ground radar, sonar. Rain, snow, dust and noise get in the way of what matters. My method, Harr-Scan, sorts the data as it comes in so the real signal stands out.

It runs alone or alongside the systems already in place, and it does not need compression to work.

Patent Pending

Harr-Scan has been patent pending since November 19th, 2026.

Where it fits

  • Lidar and radar on vehicles
  • Ground radar, finding what is buried
  • Sonar and other active sensors
Licensing inquiries are welcome, see my Contact page.

Ground Scans

Harr-Scan is Patent Pending.

Ground radar sees pipes, cables, voids and a lot of noise. Harr-Scan cuts the false flags so the real finds stand out.

Every test used public, real-world radar data. Nothing was planted or edited. Every method ran on the same data, scored the same way, on the same computer.

5x fewerfalse flags than the best opponent on cavities and voids
8x fewerfalse flags than the best opponent on pipes and cables
98.7%of cavities and voids found, tied with the best opponent
91.6%of pipes and cables found, best opponent 90.1%

Cavities and voids (79 real scans)

FoundFalse flagsSpeed per window
Harr-Scan98.7%0.037%1.54 ms
Matched hyperbola filter98.7%0.19%2.73 ms
C3 clustering96.2%0.29%0.49 ms
Harr-Scan fast mode98.7%0.36%0.20 ms

Same finds as the best opponent, 5x fewer false flags, 1.8x faster.

Pipes and cables (131 real scans)

FoundFalse flagsSpeed per window
Harr-Scan91.6%0.061%0.79 ms
C3 clustering90.1%0.49%0.23 ms
Matched hyperbola filter72.5%0.94%2.50 ms
Harr-Scan fast mode87.8%1.19%0.25 ms

More finds and 8x fewer false flags. The opponent is 3.5x faster here.

Raw radar files (Nantes test site)

Score: how much more each method flags over ground with buried targets than over documented empty ground. Higher is better.

RadarHarr-ScanBest opponentWinner
GPR 1, 400 MHz7.931.41Harr-Scan
GPR 1, 900 MHz2.334.22Opponent
GPR 2, 250 MHz1.431.19Harr-Scan
GPR 2, 500 MHz1.382.09Opponent

Two wins, two losses on raw data. Untuned, the usual opponent rules could not tell empty ground from target ground at all. Harr-Scan beat that on every radar.

Data used

  • Morocco city-survey ground radar set (Mendeley Data), 400 and 200 MHz street surveys
  • TU1208 Nantes test site (Derobert and Pajewski, 2018, CC BY 4.0), raw files with a documented empty section

Opponents

  • Matched hyperbola filter (expired patent)
  • C3 hyperbola clustering (Dou et al., 2017)
  • Layer-removed energy threshold (standard practice)

Limits, stated plainly

  • The city scans are processed images, not raw radar, with one marked item per scan.
  • Raw file results are area-level, not pipe by pipe.
  • Speeds are from one processor core. Other hardware will differ.
Want it run on your own radar data? Evaluation is available under agreement, see my Contact page.