Feature Importance in Pedestrian Intention Prediction: A Context-Aware Review

TL;DR

The research reveals the critical role of pedestrian bounding boxes and ego-vehicle speed in predicting pedestrian intentions, and potential prediction biases due to the speed feature through cross-context permutation evaluation, and proposes an alternative feature representation that enhances the contributions of input features to intention prediction.

Abstract

Recent advancements in predicting pedestrian crossing intentions for Autonomous Vehicles using Computer Vision and Deep Neural Networks are promising. However, the black-box nature of DNNs poses challenges in understanding how the model works and how input features contribute to final predictions. This lack of interpretability delimits the trust in model performance and hinders informed decisions on feature selection, representation, and model optimisation; thereby affecting the efficacy of future research in the field. To address this, we introduce Context-aware Permutation Feature Importance (CAPFI), a novel approach tailored for pedestrian intention prediction. CAPFI enables more interpretability and reliable assessments of feature importance by leveraging subdivided scenario contexts, mitigating the randomness of feature values through targeted shuffling. This aims to reduce variance and prevent biased estimations in importance scores during permutations. We divide the Pedestrian Intention Estimation (PIE) dataset into 16 comparable context sets, measure the baseline performance of five distinct neural network architectures for intention prediction in each context, and assess input feature importance using CAPFI. We observed nuanced differences among models across various contextual characteristics. The research reveals the critical role of pedestrian bounding boxes and ego-vehicle speed in predicting pedestrian intentions, and potential prediction biases due to the speed feature through cross-context permutation evaluation. We propose an alternative feature representation by considering proximity change rate for rendering dynamic pedestrian-vehicle locomotion, thereby enhancing the contributions of input features to intention prediction. These findings underscore the importance of contextual features and their diversity to develop accurate and robust intent-predictive models.

Figures (23)

• Figure 1: Pedestrian with crossing (red bounding box) and not-crossing intentions (green bounding box) in various roadway types and contexts such as crosswalk designation state, traffic-light state, and also depending on the ego-vehicle speed.
• Figure 2: The candidate pedestrian intention prediction models and their input feature. These models are distinct in architecture and fusion strategy.
• Figure 3: Histogram of proximity level of pedestrians in PIE dataset.
• Figure 4: The Probability Distribution Function (PDF) of the ego-vehicle speed in the PIE dataset.
• Figure 5: The performance of intention prediction models in distinct scenario contexts. The $S_C \cup S_{NC}$ displays the performance of models across all the crossing and not-crossing samples in the PIE dataset.