Currently we limited the maximum number of seats, which is convenient for programming and test. But in reality, it is obvious that the actual number of seats in a library is much larger. We may need to choose more advance algorithm for people detection, or to increase the number of cameras.
Universality
In the project, the usage situation is fixed by the height and angle of the camera. And the target object to be detected is human head. These limitations could be acceptable in library, but will constrain many other usages. Currently we do not have a satisfying solution for this problem, since a more universal detection algorithm is far beyond our knowledge.
Sensitivity
Now, we use the cascade classifier in OpenCV for head detection. Due to the limit of this method, our result is not very accurate. Objects with dark color can be easily be recognized as heads of people. And light condition, position of the camera, complexity of the scene and many other factors all affect the result a lot. There are many other technologies that can get more robust result, like HOG descriptor, deformable part models, tracking ,etc.
Publish
The results displayed by the webpage is adequate but not sufficient. We could add more information like history occupancy histograms, which gives users more reference to plan their schedule. Moreover, other ways could be used to publish results, like mobile application and text message subscription.