Detecting moving objects is a very important aspect of driver assistance systems (DAS). This paper handles this issue by using a vision based system mounted within the vehicle. The pipeline for both a stereoscopic and monocular approach are covered. Both approaches use image sequences and compare moving feature points over time. This sparse information is then segmented using the optimal graph-cut algorithm, by also incorporating the grey-scale images. This paper then evaluates and contrasts the two approaches to identify the accuracy and robustness of each approach. The two methods both work in real-time on normal PC hardware (Quad Core CPU).Motion is one of the major cues for human perception. Detecting moving objects is also a major issue for driver assistance systems (DAS) and road safety. The authors consider the detection of moving traffic participants to be an important step toward attention-based environment perception. This paper investigates methods and limitations of both monocular and binocular camera systems for motion detectability. It is evident that a monocular system is cheaper, uses less installation space, and suffers less decalibration issues, compared to a stereo system. However, a stereo system yields direct range measurement estimates (e.g. [5]), but the orientation between the two cameras needs to be known accurately, and decalibration can cause major issues. This paper provides insight into the difference between monocular and stereo camera performance.The motion of the ego-vehicle greatly complicates the problem of motion detection because simple background subtraction of successive images yields no result. The key idea behind our approach of detecting independently moving objects is to distinguish between motion in the images caused by the ego-motion of the ego-vehicle (static objects) and motion caused by dynamic objects in the scene. This paper presents and investigates techniques to distinguish between stationary and non-stationary 978-1-4244-2582-2/08/$25.00 c 2008 IEEE points. They are based on tracking feature points in sequential images.Section 2 presents our algorithm, starting with an investigation of motion analysis techniques, and followed by presenting segmentation of the moving objects from the static scene. In Section 3, different scenarios are presented and analysed, confirming the practicality of computer vision for the sensation and perception of motion. Differences between monocular and binocular motion detection are discussed and segmentation results for moving objects are presented. The concluding section is on future work and obtained insights.
Our AlgorithmThe proposed algorithm is able to find both rigid objects such as cars and non-rigid objects such as moving pedestrians, and it is subdivided into two main steps:Step 1. As a first result, feature points on independently moving objects are detected as moving. These features, however, are sparse and do not characterize the whole image.Step 2. In a second step, moving objects are segmented in the images using these...