Enhancement of Vehicle Detection Accuracy for Non-Chinese Traffic Contexts

[yihong1120]

Dear LeonLok,

Firstly, I would like to commend you on the exceptional work you have done with the Multi-Camera Live Object Tracking repository. The integration of Deep SORT with YOLO v4 and the comprehensive feature set for traffic counting and object tracking are impressive achievements.

Issue Overview

However, I have observed an area where the application could potentially be enhanced. As noted in your README, the DETRAC dataset primarily contains images of traffic in China, which may limit the model's ability to accurately detect and classify vehicles in different geographical contexts. For instance, the misclassification of hatchbacks as SUVs or the inability to detect taxis with varying colour schemes could be attributed to the lack of diverse training data.

Suggested Enhancement

To address this, I propose the inclusion of additional training datasets that encompass a broader variety of vehicle types and traffic scenarios from multiple countries. This could improve the model's generalisation capabilities and detection accuracy across different regions.

Potential Benefits

• Increased Accuracy: More diverse training data can lead to better model performance when deployed in non-Chinese traffic environments.
• Broader Applicability: Enhancing the model's detection capabilities can make the application more versatile and appealing to a global user base.
• Reduced Misclassification: With a more varied dataset, the likelihood of vehicle misclassification can be significantly reduced.

Conclusion

I believe that this enhancement could further solidify the application's utility and accuracy in various traffic contexts. I would be interested to hear your thoughts on this suggestion and whether it aligns with the future development plans for the project.

Thank you for considering this issue. I look forward to your response and am excited about the potential improvements to this already fantastic tool.

Best regards,
yihong1120