@chin: "the results is not what I want" does not allow to give a specific suggestion for improvements.
As in many other image processing projects I have the impression, that a much more efficient solution would be to ask a human, e.g. someone without an employment from the south states of the US, to identify the wanted properties of the image, and offer 10 cent per picture. When I compare the throughput, the development costs, the number of false positives, the reliability and maintainability, the costs of a commercial Matlab license and last but not least the social aspects and the energy consumption, the human assisted has obviously only advantages.
I'm not kidding. Some minutes ago I recommended to to perform a curve fitting by a visual estimation of a human, because this is not a scientifically hard approach. But the detection of a vacant parking space seems to be a "toy application", when there are systems which have been optimized in millions of years already for such tasks.
@Image Analyst: Do you know a service, which distributes pictures together with questions to a set of persons, who earn money with the recognition of certain properties, preferably in 24/7 real-time? I know such a service, but ironically they break gaptchas only: Instead of developing intelligent software to recognize strange combinations of letters in ugly fonts with added noise, they either send the pictures to a team of persons in Asia, or even smarter they present the same pictures online for another service and let innocent users do the work. In opposite to this, a legal application seems such straight, easy and logical, that I cannot imagine, that this has not been implemented yet.
