In a ubiquitous environment, due to continuous change of location of objects, tracking a moving object is a challenging task in intelligent IoT service provisioning. Hence, web-based IoT services can be provided by discovering and allocating only those virtual objects that are directly relevant to current location of the moving object. A semantic ontology is used to virtualize real world objects for interconnection and interoperability among virtual objects in multiple domains. Allocating relevant virtual objects requires prediction of next location of the objects through applying machine learning algorithm. Analyzing service request, prediction and real world knowledge, service request parameters are adjusted to discover relevant virtual objects dynamically as stream-variable dependency. This paper proposes an architecture that supports discovering and allocating relevant virtual objects based on context-awareness to track moving objects.