Client Success Story

    Real-Time Pallet Counting and Inventory at the Edge

    Warehouse and yard inventory counts were manual, periodic, and stale by the time they reached the WMS. Pallets moved faster than anyone could count them, and cloud video analytics were not an option across dozens of distribution sites with constrained connectivity.

    At a glance
    Real time
    Counting
    Axis IP cameras
    Sensor Layer
    None
    Video Egress
    Consumer Goods & Beverage
    Physical AI
    Anonymized — one of the world's largest beverage companies
    Physical AI
    Jetson
    Axis Communications
    Computer Vision
    Inventory
    Edge pallet detection and counting running live on NVIDIA Jetson against Axis camera streams.

    The operational problem

    Inventory accuracy in high-throughput beverage distribution degrades between counts. Pallets are staged, moved, split, and loaded continuously, so a count taken in the morning is wrong by mid-shift. Manual cycle counting is expensive and interrupts operations, barcode scanning misses anything not deliberately scanned, and streaming dozens of camera feeds to a cloud analytics service is neither affordable nor permitted at every site. The counting has to happen where the pallets are.

    What Enfuse built

    • Detection and tracking models tuned for stacked pallets, wrapped loads, and partial occlusion
    • Integration with existing Axis Communications IP cameras over RTSP — no new sensor hardware at most sites
    • Line-crossing and zone-occupancy logic for dock doors, staging lanes, and yard positions
    • Multi-camera reconciliation so a pallet seen by two cameras counts once
    • Local event buffering with store-and-forward for intermittent connectivity
    • Count events published to inventory and WMS endpoints as structured records

    Architecture

    • See — existing Axis IP cameras stream to on-site Jetson devices over RTSP
    • Understand — TensorRT-optimized detection and classification of pallets and loads on the edge
    • Locate — multi-object tracking across zones, dock doors, and staging lanes
    • Predict — dwell time and throughput signals per lane and per door
    • Act — reconciled count events pushed into inventory and WMS systems

    Hardware and software

    • NVIDIA Jetson AGX Orin / Orin NX (edge inference)
    • Axis Communications network cameras (RTSP)
    • DeepStream, TensorRT, ByteTrack-class multi-object tracking
    • On-site event buffer with store-and-forward delivery to WMS/ERP

    Deployment environment

    On-premises at distribution and manufacturing sites. Inference runs entirely on site; video never leaves the facility.

    Results

    • Pallet counts updated continuously rather than at scheduled cycle counts
    • Existing camera estate reused, avoiding a per-site sensor deployment
    • Only structured count events transit the network, keeping bandwidth and video-retention exposure flat

    Client engagement. Operational detail and site-level figures are withheld at the customer's request; the video shows the deployed edge counting pipeline.

    Next step

    Discuss this pattern against your environment

    Bring your constraints — sensors, security boundary, latency budget — and we will tell you what is realistic.