Skip to main content

Four Types of Traffic Detection (DeepStream)

This document demonstrates four traffic-related detection scenarios using NVIDIA DeepStream example applications. These demos showcase detection and counting of:

  • People
  • Vehicles
  • Traffic signs
  • Bicycles
note

These examples are basic demonstrations. For production or custom use cases, further development and configuration are required.

1. Video Detection​

This example runs detection on a video file.

Step 1: Enter the Sample Application Directory​

cd /opt/nvidia/deepstream/deepstream-7.1/sources/apps/sample_apps/deepstream-test1

Step 2: Compile the Source Code​

Refer to the README file in the same directory.

sudo make CUDA_VER=12.6
tip

On factory images, the application is often already compiled and may not require recompilation.

Step 3: Run the Demo​

Run the application using the provided configuration file:

./deepstream-test1-app dstest1_config.yml

Or run directly with a sample video:

./deepstream-test1-app ../../../../samples/streams/sample_720p.h264

Example Output​

Video Detection Result

Configuration File: dstest1_config.yml​

source:
  location: ../../../../samples/streams/sample_720p.h264

streammux:
  batch-size: 1
  batched-push-timeout: 40000
  width: 1920
  height: 1080

primary-gie:
  plugin-type: 0
  config-file-path: dstest1_pgie_config.yml

2. Real-Time Detection​

Real-time detection uses live camera input.

2.1 USB Camera Detection​

cd /opt/nvidia/deepstream/deepstream-7.1/samples/configs/deepstream-app
deepstream-app -c source1_usb_dec_infer_resnet_int8.txt

USB Camera Detection

2.2 CSI Camera Detection​

cd /opt/nvidia/deepstream/deepstream-7.1/samples/configs/deepstream-app
deepstream-app -c source1_csi_dec_infer_resnet_int8.txt

CSI Camera Detection

Summary​

  • DeepStream provides ready-to-run traffic detection demos
  • Supports video files, USB cameras, and CSI cameras
  • Configuration files control inference pipelines
  • Suitable as a starting point for intelligent traffic systems

Maintained by HemiHex for Jetson-based advanced vision and DeepStream workflows.