2nd ICAI 2022

International Conference on Automotive Industry 2022

Mladá Boleslav, Czech Republic

monitoring recordings allows for quick verification of the correctness of the algorithm and supplementing the input data set with examples of class objects that were not

correctly identified by the deep network. Figure 5: Analyse process diagram

Source: Own elaboration The premise of the application is to save frames so that the correctness of the model can be analysed. In case of missing detections, it is possible to copy the image for annotation, which is used to improve the quality of the trained model. Example frames containing incomplete detections are shown in Figure 6. on the left side of the example, the model in development failed to identify 2 empty platforms in the frame. Therefore, in the initial phase of using the software for transportation work analysis, it was assumed that the frames with detections for the ‘empty platform’ class present an inefficiency of the transportation set of 50%.

Figure 6: Sample frames saved by application

Source: Own elaboration Figure 7 shows a screenshot of a web application that indicates a summary of transportation system efficiency on a particular day. In the case in question, the potential

114