
Orbital Insight
Turning global physical data into actionable geospatial intelligence with AI-driven computer vision.
Robust Associations Multi-Pedestrian Tracking using motion and appearance information with camera-motion compensation.

BoT-SORT is a state-of-the-art multi-object tracker designed for detecting and tracking objects in a scene, maintaining unique identifiers for each. It combines motion and appearance information with camera-motion compensation and an accurate Kalman filter. BoT-SORT and BoT-SORT-ReID trackers excel in MOTChallenge datasets, achieving top ranks in MOTA, IDF1, and HOTA metrics. It leverages YOLOX and YOLOv7 for object detection and supports multi-class tracking. The architecture facilitates camera motion compensation using OpenCV's VideoStab Global Motion Estimation. Installation involves setting up a Conda environment, installing PyTorch, and using pip to install necessary packages and dependencies like ByteTrack and FastReID.
BoT-SORT is a state-of-the-art multi-object tracker designed for detecting and tracking objects in a scene, maintaining unique identifiers for each.
Explore all tools that specialize in multi-object tracking. This domain focus ensures BoT-SORT delivers optimized results for this specific requirement.
Explore all tools that specialize in pedestrian tracking. This domain focus ensures BoT-SORT delivers optimized results for this specific requirement.
Explore all tools that specialize in object detection. This domain focus ensures BoT-SORT delivers optimized results for this specific requirement.
Explore all tools that specialize in camera motion compensation. This domain focus ensures BoT-SORT delivers optimized results for this specific requirement.
Compensates for camera movements to improve tracking accuracy. It's based on OpenCV's VideoStab Global Motion Estimation.
Maintains the identity of objects even after temporary occlusions using appearance features extracted by a ReID model.
Uses a more accurate Kalman filter state vector to predict the future positions of tracked objects.
Integrates with state-of-the-art object detection models, offering flexibility and high detection accuracy.
Tracks multiple object categories simultaneously using object detectors trained on diverse datasets such as COCO.
Create a Conda environment: `conda create -n botsort_env python=3.7`
Activate the environment: `conda activate botsort_env`
Install PyTorch and torchvision: Use the appropriate command from pytorch.org for your CUDA version.
Clone the BoT-SORT repository: `git clone https://github.com/NirAharon/BoT-SORT.git`
Navigate to the BoT-SORT directory: `cd BoT-SORT`
Install the required Python packages: `pip3 install -r requirements.txt`
Install the package in develop mode: `python3 setup.py develop`
Install pycocotools: `pip3 install cython; pip3 install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'`
(Optional) Install Cython-bbox: `pip3 install cython_bbox`
(Optional) Install FAISS (CPU or GPU): `pip3 install faiss-cpu` or `pip3 install faiss-gpu`
All Set
Ready to go
Verified feedback from other users.
"Highly accurate and robust tracker, particularly effective in crowded scenes with camera motion."
Post questions, share tips, and help other users.

Turning global physical data into actionable geospatial intelligence with AI-driven computer vision.
A comprehensive set of computer vision transformations for data augmentation and manipulation in PyTorch.
Accelerate deep learning inference across Intel hardware for edge and cloud deployment.
Efficient and lightweight CNN architecture for mobile and edge devices.
Turn video into insights with BriefCam video analytics software.
Integrate powerful vision detection features into applications for image analysis and understanding.