This paper focuses on optimizing real-time intervehicle video communications from an image analysis perspective. Differently from traditional multimedia communication schemes which are typically optimized to maximize perceptual quality as perceived by human users, in this work we propose a new packet importance estimation method designed to capture the contribution of each packet to the performance of the image analysis process. Packet importance values are used to optimize video packet scheduling and retransmission in the context of ad hoc 802.11 intervehicle communications. Simulations using actual intervehicle transmission traces show PSNR gains up to 2.3 dB and 0.8 dB compared with the standard MAClayer ARQ and a distortion based optimized ARQ technique respectively, as well as up to 10% accuracy increase with the tested image analysis algorithm.