Compare LiDAR scans
Find anomalies and associate them with an event timestamp.
A PROJECT IN SPATIAL PERCEPTION FIELD NOTES / 002
Something changed.
Let’s see it.
A watchdog for the spaces you call home. The goal: let a drone or ground robot spot changes in a room and give the camera a place to look.
Step inside the experiment ↓01 / THE EXPERIMENT
A little change. A clearer picture.
A robot revisits the living room. Something is here that wasn’t here before.
POINT CLOUD / PERSPECTIVESYNTHETIC DEMO
HOME / LIVING ROOM
ROBOT PATROL · FIXED SCAN POSE
ROBOT PATROL · FIXED SCAN POSE
SCAN 0142POINTS —ΔT 10.00 s
UNFAMILIAR OBJECT ↗
↔ DRAG TO ORBIT · ARROW KEYS TO ROTATE
02 / RETURN PATROL
A bag appears in the walking area. The changed points define a region for the camera to inspect.
ROOM CHANGENEW OBJECT IN ROOM
09.00 sNEW OBJECT / 10.00 s ↗11.00 s
+ Try the reference scan, then the return patrol. Switch to 2D to see where the new object lands.
ILLUSTRATION, NOT LIVE SENSOR DATA02 / FROM SPACE TO SIGHTGive the camera
Give the camera
a place to look.
A home patrol can tell you where a room changed. A matching camera image gives a model the context to inspect it. A change is a cue to investigate, not proof of a threat.
SCAN COMPARISON
The starting signal is an anomaly: a set of changed 3D points and the time they were observed. Haymaker’s projection module accepts those points directly or selects them with a boolean mask.
↳03 / THE CONNECTIONTwo sensors.
Two sensors.
One shared view.
Calibration is the bridge. Rotate and translate a LiDAR point into the camera’s coordinate system, then use the camera’s intrinsics to find its pixel.
pcamera = R · pLiDAR + T
Illustrative translation only. Real hardware needs measured calibration. ONE POINT, TWO COORDINATE SYSTEMS
PLANNED
Bring 3D into 2D
Timestamp matching, calibrated projection, visible regions, and an ML callback.
Add visual understanding
Connect an object detection model to the prepared camera region.