Image object detection is a task valuable for both theoretical research and practical usage, and is served to many fields of defense as well as civil economic construction. The traditional linear and statistical theories can no longer satisfy the needs of advanced object detection. Mathematical morphology is a nonlinear theory for image and signal analysis and processing, which can estimate many information of the geometrical structure in the signal and concurs with our instinctive perceptual system. For this reason, it has been paid more and more attentions and developed rapidly. In this paper, main works needed by an object detection system are deeply researched based on a thorough investigation of the basic theories of mathematical morphology and its current usages in object detection. The main works include image feature extraction, object modeling, and detection algorithm.For image feature extraction, two feature extraction methods based on mathematical morphology are proposed. One is a morphological edge extraction algorithm by using multiple structure elements, which can achieve both the edge magnitude and orientation, and is precise, quick, and immune to noise. The other one is a morphological corner extraction algorithm based on thehit-number, which is suitable to both the binary and the gray-scale images. The algorithm has high detection accuracy, strong noise restraining ability, and is simple in calculation and easy to be realized on hardware.For shape modeling and detection, a shape representation algorithm based on morphological shape decomposition is proposed. Based on the decomposition scheme, we construct a fuzzy attributed relational graph by learning from training examples as the shape model. A hierarchical part-based shape detection algorithm is developed based on such a model. The shape representation algorithm concurs with our instinctive perceptual system, and is suitable for constructing the detection algorithm. Compared to other shape matching algorithms, our shape modeling and detection algorithm has distinguishing features. Our part-based and hierarchical strategy finds each part of the shape sequentially, rather than extracting all shapes from the image to match with the possible shape models. It can improve the accuracy and speed the detection processing, and can help to reduce or eliminate the effects of background clutters, partial occlusions and inner rotation in the object. It is also easy in parallel form realization.For multivalued image object detection, a strategy called conditional flooding is proposed to develop a general multivalued morphological object detection framework, in which different object detection tasks are unified. Two algorithms are proposed to implement the conditional flooding, including the local conditional flooding and the global