Fig. 5. NASA TLX user workload
The results of the evaluation shown in Figure 5 reveal a trend that the workload is lower for the zoomed videos. The TLX workload values show that the zooming approach better fits the sports videos. For the talk videos the workload is reduced by 26 % using zooming and for the sport videos the workload is reduced by 49 %. The results lend support that zooming is a good solution to improve mobile user experience for mobile video. The analysis of the video browsing results shows a larger improvement of mobile user experience with much lower workload than the traditional video. Compared to the zooming approach, the type of video content has a big impact on the workload. The workload for the documentary film videos is reduced by 67 % and for the sports video by 64 %.
B. Cloud evaluation
One of the main advantages of the cloud infrastructure for the intelligent video processing services is the accelerated processing/adaptation time. To evaluate the MVCS cloud part three videos with different lengths were chosen. The first video is processed as one chunk on one instance, the second one as three chunks on three instances and finally the third one as five chunks on five instances. Amazon EC2 small instances were used as cloud infrastructure. For every task the processing time is measured. For the chunk-based approach this means the process from splitting the video into chunks, processing them and merging them again. The MVCS cloud evaluation proves that splitting the video into chunks instead of processing it as a single file is a good solution to deliver faster video processing for near real-time delivery.
Fig. 6. Cloud processing time comparison
Figure 6 presents a comparison of all cloud evaluation results. Here again, we see that the processing time for the single file approach is increasing with the length of the video. For the chunk-based approach we see that the processing time is increasing very slightly. The small increase can be explained that merging the chunks to one file takes a little more time for every additional chunk. Nevertheless, our system has also limitation. The user study showed that zooming has to be improved as some users lost the context of the video due to the zooming. The number of evaluation participants and types was limited. A real-world evaluation over a longer period is
needed to assess the perceptual quality of the delivered video streams.