INVESTIGATION INTO ALTERNATIVE AND AUTOMATED RECORDING OF BCS IN COWS.

Automated BCS enables the owners to score their cows more often and more consistently compared to manual methods. A common approach to automated BCS scoring is using a regression-based model. This involves extracting features from the cows digitally and then fitting these features to the manually labelled BCS values through various regression techniques. These regression based models can then be used to predict the BCS of a never before seen cow. A newer approach to automated BCS is using Convolutional Neural Networks (CNN). The system uses a 3D scan of the cow’s back. A similar approach was used here, albeit with multiple 3D cameras.

The camera frame was deployed at the University of Pretoria experimental farm. The frame was placed over the crush above the weighing platform. Three 3D cameras were used to capture the cow body from different angles – top, rear, angled – in order to obtain enough body information. During the course of the investigation several changes were made to the system and the data collection procedure to ensure the process went smoothly and as much of the collected data was accurate and usable as possible. These changes were made due to what was learned and experienced in the first and second data collection sessions. The first change involved some upgrades to the system. The three Jetson Nano computers, which are used to capture and send the depth frames to the server, were set to synchronise their time to that of the server (a laptop in this case). The captured depth frames now also include this synchronised time in the file name. Another small change to the data collection procedure was the addition of a normal RGB camera which was used to record the cows as they entered the sorting gate. The camera was mounted on a tripod near the sorting gate and recorded the entire data collection session. This was done so that in the case where the data collection system has an error and misses a cow or the same perhaps happens to the farm’s tag reader, then the video footage would help understand how many cows passed through the frame as well as when they passed through.

The procedure thus far appears promising and was tested also in a commercial operation with more than 100 cows. The investigation continues.