Related Experiment Videos
Continuous Tracking and Recognition of Small Objects in Video Streams Based on YOLO and Spatio-Temporal Contextual
Chengyuan Pang1,2,3, Zongpu Li1,2,3, Le Ru1,2,3
1Equipment Management and Unmanned Aerial Vehicle Engineering School, Air Force Engineering University, Xi'an 710051, China.
Sensors (Basel, Switzerland)
|July 28, 2026
Summary
This study introduces a novel method for tracking small objects in videos using an improved YOLOv8 model and a spatio-temporal context memory network. The approach enhances feature extraction and fusion, achieving over 93% success in continuous small object tracking and recognition.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Small objects in video streams present challenges for tracking due to limited visual information.
- Existing methods struggle with continuous identification and recognition of small objects.
Purpose of the Study:
- To develop a robust method for continuous tracking and recognition of small objects in video streams.
- To address the limitations of current techniques in handling small object data.
Main Methods:
- An improved YOLOv8 backbone network with a wavelet pooling module for multi-scale feature extraction.
- A mixed attention module to enhance spatially significant features.
- A spatio-temporal context memory network and bidirectional feature pyramid for multi-scale spatio-temporal feature fusion.
Main Results:
- The proposed method effectively extracts spatiotemporal features from small objects in video datasets.
- Achieved a success rate exceeding 0.93 for continuous tracking and recognition under varying occlusion levels.
Conclusions:
- The developed method demonstrates significant improvements in small object tracking and recognition.
- It offers a viable solution for scenarios with numerous small objects and partial occlusions.
