Automated Multiple View Inspection Based on Uncalibrated Image Sequences

Abstract

Abstract. The Automated Multiple View Inspection (AMVI) has been recently developed for automated defect detection of manufactured objects. The approach detects defects by analysing image sequences in two steps. In the first step, potential defects are automatically identified in each image of the sequence. In the second step, the potential defects are tracked in the sequence. The key idea of this strategy is that only the existing defects (and not the false detections) can be successfully tracked in the image sequence because they are located in positions dictated by the motion of the test object. The AMVI strategy was successfully implemented for calibrated image sequences. However, it is not simple to implement it in industrial environments because the calibration process is a difficult task and unstable. In order to avoid the mentioned disadvantages, in this paper we propose a new AMVI strategy based on the tracking of potential detects in uncalibrated image sequences. Our approach tracks the potential defects based on a motion model estimated from the image sequence self. Thus, we obtain a motion model by matching structure points of the images. We show in our experimental results on aluminium die castings that the detection is promising in uncalibrated images by detecting 92.3% of all existing defects with only 0.33 false alarms per image.

Automated multiple view inspection of metal castings

“…In addition, the vibrations of the imaging system induce inaccuracies in the estimated parameters of the model, i.e., the calibration is not stable and the computer vision system must be calibrated periodically (off-line) in order to avoid uncertainty. For these reasons, we developed approaches based on the tracking of potential detects in two views 17,18 and in three views 18 using uncalibrated image sequences, in which it was not necessary to calibrate the imaging system. This new approaches track the potential defects based on a motion model estimated from the image sequence itself.…”

“…The structure points are matched using B-Spline curves and correlated curve sections of the structure. 17,18 Using RANSAC 22 the matched structure points are employed to estimated the bifocal and trifocal tensors required for the multiple view analysis. In this sense, we do not calibrate the image sequence, we only estimate the bifocal and trifocal tensors required for the tracking.…”

Automatic Multiple Visual Inspection on Non-calibrated Image Sequence with Intermediate Classifier Block

“…AMVI is currently a useful tool and a powerful alternative for examining complex objects. It provides two independent approaches: those based on the calibration of a 3D→2D transfer function in the multiple view projection [6], and those based on the estimation of the motion of the control points in correspondence for pairs [7] and triplets of views [8] without prior calibration. A brief description of each is given below.…”

“…To counteract the difficulties mentioned above, in recent years a new methodology has been developed to detect flaws automatically making use of the potential of multiple views called Automatic Multiple View Inspection [6,7,8,9] (AMVI). The main objective of AMVI is to exploit the redundancy of information in multiple views that have corresponding parts of the object that is being analyzed, so the information captured from different viewpoints can reinforce the diagnosis made with a single image.…”

“…Unfortunately, the calibration process is difficult to carry out in industrial environments due to the vibrations and random movements that vary in time and are not considered in the original estimated transfer function. An alternative method for carrying out the AMVI strategy in non-calibrated sequences was presented in [7] for sequences with two images, and in [8] for sequence with three images. In order to achieve an adequate performance, the number of false alarms to be tracked must be reduced.…”