Vehicle Detection in Stereoscopic Images Using Symmetry-Based Approach

This study introduces a unique symmetry-based approach for vehicle detection in stereoscopic images, a critical advancement for driver assistance systems. Distinct from existing techniques, this method innovatively combines the Canny operator with corner points to create a more precise and computationally efficient vehicle detection system. The method uses the canny operator and corner points to identify contours and then employs symmetry maps and moment calculations to link these contour points. The method was tested using the KITTI stereo dataset, which contains real-world driving scenarios, and compared with established methods like the Harris corner detector and Scale-Invariant Feature Transform (SIFT). The new approach showed superior accuracy, with an average precision (AP) of 92.3% and average recall (AR) of 87.5%, outperforming the Harris detector (AP = 85.7%, AR = 81.2%) and SIFT (AP = 89.1%, AR = 84.6%). Additionally, it was more computationally efficient, processing frames in 0.034 s on average, faster than both the Harris detector (0.057 s) and SIFT (0.049 s). Further validation was done using the Middlebury Stereo Dataset and the Karlsruhe Urban Stereo Dataset. The method continued to demonstrate high performance, achieving AP values of 87.2% and 84.6% on these datasets, respectively, again outperforming the Harris detector and SIFT. This indicates the method’s robustness and effectiveness in various scenarios.