Vehicle detection and recognition approach in smart surveillance system: A comparative analysis

Vehicle detection and recognition are one of the main research areas of computer vision and image processing. In recent years, the recognition of vehicles from video clips have been a critical component of intelligent transportation systems (IRS's). It is used for the purpose of detecting and monitoring vehicles, capturing their violations, and regulating traffic. This paper addresses the key issue of vehicle detection and recognition, for these two complex datasets, VRiV and UCSD are considered to segregate between vehicle and non-vehicle objects using video analysis. For data analysis we apply basic preprocessing, which includes frame conversion, background subtraction, noise reduction, and resizing. After that, we implement region-of-interest (ROI) extraction and change detection techniques to optimize the ROI. Next step is extraction of main attributes, such as motion direction and motion angle and apply data normalization to improve the equity of the model training. The study assess the Long Short-Term Memory (LSTM) model's efficacy on the data using benchmark metrics such as confusion matrix, precision, recall, accuracy, and F1-score. The proposed system attain an accuracy of 81 % on the VRiV dataset and 83 % on the UCSD datasets which shows the robust performance in terms of recognition rate.