In recent years, the computer vision field has had significant success on the object detection task in images. However, current state-of-the-art approaches are computationally demanding and not well suited for efficient processing of videos. This Thesis focuses on methods that efficiently detect moving objects in a video and can be scaled for such applications as highway or people monitoring using inexpensive devices with limited computational power. The literature review discusses the different object detection methods and identifies which of the approaches are more accurate, which are robust in the changing environment, and which are computationally efficient. Based on the literature review, the Thesis develops novel video-based moving object detectors that process only a line of pixels in a frame. The first developed method – Intervals on a Virtual Detection Line (IoVDL) – exceeds other existing efficient methods with an interval approach, which allows use of the detection line in more varied scenes than conventional region-based detectors. The versatility of the interval approach is further demonstrated by developing an extended IoVDL. Using several detection lines in a frame, the extended IoVDL is capable of tracking and characterizing objects, while being computationally efficient. Another proposed method – Recurrent Neural Network-based Virtual Detection Line (RNN- VDL) – combines the efficiency of the detection line approach and the versatility of machine learning. This method requires specific training data, so this Thesis explores and develops novel data labeling methods. The tests of developed and implemented methods include the comparison of vehicle counting by IoVDL and RNN-VDL, the measurement of classification accuracy of the extended IoVDL, and the measurement of the computational efficiency of the proposed methods. The versatility of RNN-VDL is tested by retraining the neural network-based method and using it for people counting in a video. The Thesis proves four statements and, as a result, efficient and adaptable methods for detection of moving objects in a video are developed. It consists of 132 pages, 37 figures, 4 tables, 2 algorithms, 111 sources of literature and 6 appendices. The Doctoral Thesis has been developed in the Institute of Electronics and Computer Science (EDI).