yzfzzz/depth-detect

mono / stereo depth estimation + yolo object detection with deployment on Jeston nano / TX2 / GeForce by TensorRT

What it solves

It provides a high-performance framework for combining object detection and depth estimation in real-time. By fusing these two capabilities, the system can not only identify objects but also estimate their distance and motion state (velocity and acceleration), which is critical for applications like obstacle avoidance or robotic perception.

How it works

The framework uses a C++ and TensorRT-based pipeline to run two models in parallel: a detection model (YOLO) and a depth estimation model (Depth-Anything or Lite-Mono). It supports both synchronous and asynchronous (GPU overlap) execution to maximize throughput. Once objects are detected, the system uses BYTETracker for multi-object tracking and a Kalman filter to smooth the motion state and estimate distance changes based on the depth map.

Who it’s for

It is designed for developers working on computer vision systems for x86 Linux or NVIDIA Jetson embedded platforms who need low-latency, real-time spatial awareness.

Highlights

  • Parallel Inference: Supports synchronous and asynchronous pipelines for simultaneous depth and detection processing.
  • Hardware Acceleration: Utilizes CUDA for pre- and post-processing (Resize, Normalize, NMS) to reduce CPU overhead.
  • Motion Tracking: Integrates BYTETracker and Kalman filtering to track objects and estimate their movement dynamics.
  • Cross-Platform: Compatible with both desktop GPUs (RTX series) and embedded devices (Jetson TX2).

Related

  • Project
    cyrusbehr/YOLOv8-TensorRT-CPP

    A C++ implementation of YOLOv8 using TensorRT for high-performance GPU inference, supporting object detection, semantic segmentation, and pose estimation.

  • Project
    hank-ai/darknet

    An open-source neural network framework written in C/C++ and CUDA that powers the YOLO real-time object detection system.

  • Project
    triple-mu/YOLOv8-TensorRT

    A high-performance TensorRT acceleration wrapper for YOLOv8, enabling fast inference for detection, segmentation, pose, OBB, and classification in both Python and C++.

  • Project
    Geekgineer/YOLOs-CPP

    A production-ready C++ inference engine that provides a unified API for the entire YOLO family, supporting tasks from object detection to metric depth estimation.

  • Project
    ultralytics/ultralytics

    A comprehensive computer vision framework featuring YOLO models for real-time object detection, segmentation, classification, and pose estimation.