Object detection has been widely used in intelligent video surveillance, robot navigation, industrial detection, and other fields. Object detection can effectively reduce the consumption of human capital and has important practical significance. However, the existing object detection methods lack adaptability to the application environment and need to obtain all images classes at one time to train the model in a static setting, and do not support the incremental learning. Therefore, we propose a novel Class-Incremental Object Detection (CIOD) framework based on deep learning. CIOD divides object detection into two stages:object candidate box generation and selection. In the first stage, we improved the traditional OpenCV cascaded classifier to adapt to class-incremental learning while maintaining accuracy. In the second stage, we use example sets and prototype vectors to construct a class increment-based classifier to identify the generated object candidate box. We verify the effectiveness of the proposed method in terms of object detection effect, efficiency, and memory capacity. Experimental results show that CIOD can detect object in a class-incremental manner and can control the memory capacity not to increase with the number of newly increased classes.