The JSCZ vehicle algorithm box is an edge-AI processing unit for vehicle thermal night-vision systems. It runs person/vehicle recognition on the thermal camera feed and streams structured detection results over a simple UART protocol, letting host devices and ADAS/assisted-driving displays act on alerts without heavy onboard image processing.
Real-time person & vehicle recognition
● on the thermal night-vision stream
Structured one-way UART output
● (115200 bps, 8N1) device transmits only when a target is detected
Distance reporting
● with alarm levels:
○ Persons incl. cyclists/e-scooters/motorcycles/three-wheelers; 3-level proximity alarm (≥60m no alarm; 40-60m level 1; 20-40m level 2; <20m level 3) with per-frame distance in meters
○ Motor vehicles detected with distance value and other flags
Bounding-box coordinates
● for every detected target on a 1280×719 image plane direct draw-able data
● Designed for rail, mine vehicles, driverless logistics trucks, ports & harbors
● Pairs with JSCZ640 / JSCZ-series thermal cameras
● Person frame:
$p, Y1,X1,Y2,X2, level+distance, frameID**checksum
● Vehicle frame:
$c, Y1,X1,Y2,X2, flag+distance, frameID**checksum
● Coordinates normalized to a 1280×719 plane for all camera resolutions
● ASCII checksum; transmits only on detection (no continuous stream)
Specs sourced from JSCZ vehicle thermal night-vision person/vehicle recognition algorithm box comms protocol V1.4.7 full protocol document available on request.