Given a set of spatial objects of different features (e.g., mall, hospital) and a spatial relation (e.g., geographic proximity), the problem of local co-location pattern detection (LCPD) pairs co-location patterns and localities such that the co-location patterns tend to exist inside the paired localities. A co-location pattern is a set of spatial features, the objects of which are often related to each other. Local co-location patterns are common in many fields, such as public security, and public health. For example, assault crimes and drunk driving events co-locate near bars. The problem is computationally challenging because of the exponential number of potential co-location patterns and candidate localities. The related work applies data-unaware or clustering heuristics to partition the study area, which results in incomplete enumeration of possible localities. In this study, we formally defined the LCPD problem where the candidate locality was defined using minimum orthogonal bounding rectangles (MOBRs). Then, we proposed a Quadruplet & Grid Filter-Refine (QGFR) algorithm that leveraged an MOBR enumeration lemma, and a novel upper bound on the participation index to efficiently prune the search space. The experimental evaluation showed that the QGFR algorithm reduced the computation cost substantially. One case study using the North American Atlas-Hydrography and U.S. Major City Datasets was conducted to discover local co-location patterns which would be missed if the entire dataset was analyzed or methods proposed by the related work were applied.