Vision-Based Activity Classification of Excavators by Bidirectional LSTM

Advancements in deep learning and vision-based activity recognition development have significantly improved the safety, continuous monitoring, productivity, cost of earthwork site. The construction industry has adopted CNN RNN models to classify different activities equipment automate operations. However, currently available methods based on visual information current frames. To date, adjacent frames not been simultaneously examined recognize industry. This paper proposes a novel methodology excavator by processing video frame. follows CNN-BiLSTM standard pipeline for recognition. First, pre-trained model extracted sequential pattern features from Then BiLSTM classified analyzing output convolutional neural network. forward backward LSTM layers stacked help algorithm compute considering previous upcoming frames’ information. Experimental results shown average precision recall be 87.5% 88.52%, respectively.