Real-time Vehicle Detection, Classification & Tracing Using State-Of-The-Art Dataset

In Today's era, real-time Detection of Vehicles has become highly challenging because, mainly every coming year, various new Vehicles of different models, different skeleton structures, and varied colours are produced. These vast differences make Real-time detections of the Vehicles Approach invalid and less accurate, thus causing an imbalance in the requirement. Furthermore, the real-time Detection of Vehicle Approach is not well trained on Day-Night images, further increasing a loss of response in the frame of time needed for Real-time Detection. The previously developed approach does not process or enhance the Day-Night Frames, which bring down the response time, and responding accuracy has been seen to go drastically down. We solely aim to improve the automobile classification accuracy at Daytime-Night-time & Night-Time and provide experimental results. We work on classifying the Vehicle type better to give a further matured accuracy after working on Vast Classified Vehicle Types. Detection of the object under Night-time conditions such as soft illumination, an excessive amount of Glare, non-visible contrast makes the process of classifying object during Night-time a challenging task. The development is distributed into several segments/steps, respectively casing various prerequisites and stipulations for the submission's efficacious accomplishment and disposition. We conducted a unique experiment on our Self-Generated Dataset (VEHICLE_BRL_DTU). This unique and novel experimentation helped us find that the accuracy of our approach is very high if the Real-time Videography per frame rate is run through our enhancing application, thus further pushed to our application.