Object Tracking with Kalman Filter and YOLO Inference

Overview

This project demonstrates object tracking using the Kalman Filter, integrated with YOLO for object detection. It leverages NVIDIA's TensorRT for fast inference of pre-trained deep learning models, enabling real-time object detection and tracking.

In this implementation:

1. YOLO performs object detection on frames of a video.
2. The Kalman Filter is applied for object tracking across frames.

This project is based on the original TensorRT C++ API for efficient TensorRT usage.

Prerequisites

In addition to the prerequisites listed in the TensorRT C++ API, the following are required:

1. GTest: Used for unit testing the Kalman Filter implementation. Install Google Test for running tests.

Building the Project

• mkdir build
• cd build
• cmake ..
• make -j$(nproc)

Running the Executables

To run the executables, use the following commands:

1. Using camera as input: ./detect_object_video --model /path/to/your/onnx/model.onnx --input 0

2. Using video as input: ./detect_object_video --model /path/to/your/onnx/model.onnx --input /path/to/videofile

3. To record the result: ./detect_object_video --model /path/to/your/onnx/model.onnx --input /path/to/videofile --output validmp4filename The video will be saved in the outputs folder.