Tracking and classification of multiple objects in multibeam sector scan sonar image sequences

Multi-beam forward looking sector scan sonars are commonly used as obstacle avoidance and relative navigation sensors on unmanned underwater vehicles. Their key characteristic is a fast update rate (e.g. 12 Hz at 10 metres range). This offers opportunity to exploit temporal as well as spatial correlation in automatic processing of the data. We present the approach to object segmentation, tracking and classification, exploiting both inter and intra frame processing. Using optical flow motion estimation, coupled to a tree structure allowing object tracks to be revised, we have demonstrated good tracking performance, with prediction errors of between 10 and 50 cm (1-5% of scan range). Supervised object classification has demonstrated errors of 1 to 2 % using non-noisy images. With realistic sensor noise, classification of up to 100% was achieved with signal-to-noise ratio between 7.6 and 9.5 dB