Deep Learning for Vehicle Tracking and Wheel Detection

Deep learning-based vehicle tracking and segmentation using SiamFC, DeepLabV3+, and Mask R-CNN for speed analysis.

The Vehicle Speed Analysis System is designed to automatically analyze the speed of vehicles in video footage. The project includes two main modules:

1. Module 1 - Object Tracking:This module uses Siamese Fully Convolutional Networks (SiamFC) to track a target vehicle across frames. The system is enhanced with a Kalman filter to improve tracking accuracy under varying conditions such as occlusions, low resolution, and shape changes. The implementation uses MATLAB's Deep Learning Toolbox to integrate pre-trained models for efficient tracking.
2. Module 2 - Instance Segmentation:This module focuses on precise segmentation of vehicle parts using DeepLabV3+ for semantic segmentation and Mask R-CNN for instance segmentation. The methodology is tailored to overcome challenges like scale variations and low resolution. A custom dataset, annotated for segmentation tasks, is used for model training and validation.

Key project highlights include:

• Deployment of deep learning models with MATLAB for tracking and segmentation.
• Comparative analysis of DeepLabV3+ and Mask R-CNN performance.
• Pretrained models for Module 2 can be downloaded from the following links:

- Download Pretrained Model 1(MaskRCNN)

- Download Pretrained Model 2(DeepLabv3)

This system demonstrates the potential of MATLAB's deep learning frameworks to solve complex computer vision tasks effectively.