bytetrack-counter-go

Edge-computing live video analytics counter for industrial poultry processing. Connects to an RTSP camera, runs YOLO object detection on a Rockchip RKNN NPU, tracks objects with ByteTrack, and counts chickens ("ayam") crossing a virtual line. Closes counting batches when a cutting board ("talenan") crosses the line. Persists batch data to SQLite.

Architecture

RTSP Camera --> gocv.VideoCapture --> YOLO (RKNN NPU) --> ByteTrack
                                                              |
                                                     Line-Crossing Logic
                                                              |
                                              +-----> BatchStore (SQLite)
                                              |
                                         HUD Drawing --> Video Output / Live JPEG

Requirements

• Rockchip RK3588 (or compatible) with RKNN NPU
• librknnrt (Rockchip RKNN Runtime library, >=1.6.0)
• OpenCV 4.x with GoCV bindings
• Go 1.23+
• CGo enabled (required for RKNN and GoCV)

Installing GoCV / OpenCV

On RK3588 (Armbian/Ubuntu):

sudo apt install -y libopencv-dev pkg-config
go get -u gocv.io/x/gocv

Installing librknnrt

# Copy librknnrt.so to the device (obtained from Rockchip SDK)
sudo cp librknnrt.so /usr/lib/aarch64-linux-gnu/
sudo ldconfig

Verify:

ls /usr/lib/aarch64-linux-gnu/librknnrt.so
# Should also have: librknn_rk3588.so or similar NPU driver

Compiling

Clone and build:

git clone https://git.proit.id/dsutanto/bytetrack-counter-go.git
cd bytetrack-counter-go
CGO_ENABLED=1 go build -o counter ./cmd/counter/

The -o counter flag places the binary at ./counter. For production, copy it to /opt/bytetrack-counter-go/counter.

To verify the binary links correctly:

ldd ./counter | grep rknn
# Should show librknnrt.so => /usr/lib/aarch64-linux-gnu/librknnrt.so

Cross-compiling from amd64

CGO_ENABLED=1 GOOS=linux GOARCH=arm64 \
  CGO_LDFLAGS="-L/path/to/rknn-libs -lrknnrt" \
  go build -o counter ./cmd/counter/

Configuration

All configuration is via environment variables. Copy the example and edit:

cp config.env.example .env
nano .env

Key settings:

Variable	Default	Description
SOURCE	—	RTSP URL or video file path (required)
MODEL_PATH	—	Path to .rknn YOLO model (required)
OUTPUT_DIR	/opt/jetson-counter	Output directory for DB, state, CSV, video
IMGSZ	320	YOLO input image size (must match model)
CONF	0.3	Detection confidence threshold
CORE_MASK	1	RKNN NPU cores: 1=core0, 2=core1, 3=both, 7=all three
NUM_CLASSES	2	Number of YOLO classes
CLASS_AYAM	ayam	Object class to count
CLASS_TALENAN	talenan	Object class to close batches
LINE_X_FRAC	0.5	Counting line X position (fraction of frame width)
CROSS_DIRECTION	rtl	Crossing direction: rtl, ltr, or both
DAILY_CUTOFF_TIME	20:00	Time to roll over to next counting day
BATCH_TIMEOUT_SECONDS	300	Inactivity timeout to close a batch
MIN_OBJECT_PER_BATCH	60	Minimum chickens per batch
OPENCV_FFMPEG_CAPTURE_OPTIONS	—	FFmpeg options for low-latency RTSP

See config.env.example for the full list.

Running

Interactive (foreground)

set -a && source .env && set +a
./counter

systemd (production)

# Install binary and config
sudo mkdir -p /opt/bytetrack-counter-go
sudo cp counter /opt/bytetrack-counter-go/
sudo cp .env /opt/bytetrack-counter-go/
sudo cp config.env.example /opt/bytetrack-counter-go/

# Install and start service
sudo cp bytetrack-counter-go.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable bytetrack-counter-go
sudo systemctl start bytetrack-counter-go

# Monitor
sudo journalctl -u bytetrack-counter-go -f

Directory Structure

.
├── cmd/counter/main.go          # Application entrypoint
├── pkg/
│   ├── batchstore/store.go      # SQLite batch persistence + JSON state
│   ├── bytetrack/tracker.go     # ByteTrack multi-object tracker
│   ├── config/config.go         # Env-based configuration loader
│   ├── drawing/drawing.go       # OpenCV HUD drawing helpers
│   ├── kalman/kalman.go         # 8-state Kalman filter
│   ├── rknn/rknn.go             # CGo bridge to librknnrt
│   └── yolo/yolo.go             # YOLO detector (RKNN inference + NMS)
├── internal/
│   ├── iou/iou.go               # IoU computation + greedy matching
│   └── nms/nms.go               # Non-maximum suppression
├── config.env.example           # Example configuration
├── bytetrack-counter-go.service # systemd unit file
├── go.mod
└── go.sum