Abstract:

This paper presents a full-field thermal displacement inversion and dynamic disparity compensation method to address the disparity drift issue in mass-produced automotive-grade stereo cameras induced by thermal expansion and rigid-body displacement of image sensors under wide-temperature-range operating conditions. The proposed method leverages a high-precision static speckle pattern and digital image correlation (DIC) technique to achieve quantitative characterization of the non-uniform horizontal thermal deformation fields of the left and right image sensors. On this basis, a dynamic disparity compensation model is established, integrating a temperature-dependent quadratic polynomial translation term and a radial thermal expansion correction term. Experiments were conducted by heating the camera from 35 °C to 70 °C while synchronously acquiring speckle images, enabling quantitative extraction of the full-field displacement evolution of the sensors in both X and Y directions. The results reveal that the thermally induced sensor displacement consists of two superimposed components: a global rigid-body translation and a radial expansion centered at the image origin, with a maximum translation of approximately 0.5 pixels. The compensation model built upon these findings effectively corrects the thermal-induced disparity bias. In simulated on-road ranging tests, without compensation, the maximum ranging errors at 5 m for the center and edge corner points were 2.44% and 3.02%, respectively, and reached 4.87% and 5.42% at 10 m. After applying the proposed compensation model, these errors were reduced to 0.68% and 0.92% at 5 m, and to 1.43% and 2.47% at 10 m, corresponding to an overall ranging accuracy improvement of 65%-72%. This study elucidates the coupling mechanism between thermal strain and stereo ranging errors, offering an engineering solution for high-robustness ranging of vehicle-mounted stereo vision systems under complex temperature-varying conditions.

Key words: vehicle-mounted stereo camera, temperature compensation, digital image correlation (DIC), thermal displacement, embedded real-time compensation, depth estimation stability